natas_hw
/
cassandra
mirror of https://github.com/apache/cassandra
0
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Improve Observability: 

- Track all active Coordinations
 - Refactor Replica/Coordinator metrics and report Coordinator exhausted/preempted/timeout
 - DurabilityQueue metrics and visibility
Also Fix:
 - WaitingState can get cause distributed stall when asked to wait for CanApply if not yet PreCommitted; track separate querying state and advance this to the next achievable state rather than the desired final state
 - Stalled coordinators should not prevent recovery
 - Edge case with fetch unable to make progress when pre-bootstrap and all peers have GC'd
 - Dependency initialisation for sync points across certain ownership changes
 - SyncPoint propagation may not include all of the epochs required on the receiving node for ranges they have lost but not closed, and receiving node does not validate them
 - Stable tracker accounting with LocalExecute
 - Do not prune non-durable APPLIED as must be reported in dependencies until durably applied (so as not to break recovery)
 - Ensure we cannot race with replies when initiating Coordination
 - ProgressLog does not guarantee to clear home or waiting states when erased or invalidated by compaction
 - WaitingState on non-home shard cannot guarantee progress once home shard is Erased
 - WaitingOnSync handles retired ranges incorrectly
Also Improve:
 - Standardise failure accounting, use null to represent single reply timeouts
 - BurnTest record/replay to/from file

patch by Benedict; reviewed by Alex Petrov for CASSANDRA-20878
This commit is contained in:
Benedict Elliott Smith 2025-08-02 16:44:55 +01:00
parent eeb227be2f
commit a813a51f7c
77 changed files with 2572 additions and 724 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
modules/accord

@ -1 +1 @@
Subproject commit 5cbe8d62f15cc7d66af2c7686f14b8fa52b1d35b
Subproject commit 48061a521cde7a085fe7a5503023c8ef8b687a89

128
src/java/org/apache/cassandra/db/virtual/AccordDebugKeyspace.java
View File

@ -41,6 +41,13 @@ import javax.annotation.Nullable;
import com.google.common.collect.BoundType;
import com.google.common.collect.Range;
import com.google.common.collect.Sets;
import accord.coordinate.AbstractCoordination;
import accord.coordinate.Coordination;
import accord.coordinate.Coordinations;
import accord.coordinate.PrepareRecovery;
import accord.coordinate.tracking.AbstractTracker;
import accord.utils.SortedListMap;
import org.apache.cassandra.cql3.Operator;
import org.apache.cassandra.db.EmptyIterators;
import org.apache.cassandra.db.filter.ClusteringIndexFilter;
@ -57,7 +64,6 @@ import accord.api.TraceEventType;
import accord.coordinate.FetchData;
import accord.coordinate.FetchRoute;
import accord.coordinate.MaybeRecover;
import accord.coordinate.RecoverWithRoute;
import accord.impl.CommandChange;
import accord.impl.progresslog.DefaultProgressLog;
import accord.impl.progresslog.TxnStateKind;
@ -151,6 +157,8 @@ import static org.apache.cassandra.utils.MonotonicClock.Global.approxTime;
public class AccordDebugKeyspace extends VirtualKeyspace
{
public static final String COORDINATIONS = "coordinations";
public static final String EXECUTORS = "executors";
public static final String COMMANDS_FOR_KEY = "commands_for_key";
public static final String COMMANDS_FOR_KEY_UNMANAGED = "commands_for_key_unmanaged";
public static final String DURABILITY_SERVICE = "durability_service";
@ -173,6 +181,8 @@ public class AccordDebugKeyspace extends VirtualKeyspace
private AccordDebugKeyspace()
{
super(VIRTUAL_ACCORD_DEBUG, List.of(
new ExecutorsTable(),
new CoordinationsTable(),
new CommandsForKeyTable(),
new CommandsForKeyUnmanagedTable(),
new DurabilityServiceTable(),
@ -192,6 +202,118 @@ public class AccordDebugKeyspace extends VirtualKeyspace
));
}
// TODO (desired): human readable packed key tracker (but requires loading Txn, so might be preferable to only do conditionally)
public static final class ExecutorsTable extends AbstractVirtualTable
{
private ExecutorsTable()
{
super(parse(VIRTUAL_ACCORD_DEBUG, EXECUTORS,
"Accord Executor State",
"CREATE TABLE %s (\n" +
" executor_id int,\n" +
" status text,\n" +
" position int,\n" +
" unique_position int,\n" +
" description text,\n" +
" command_store_id int,\n" +
" txn_id 'TxnIdUtf8Type',\n" +
" txn_id_additional 'TxnIdUtf8Type',\n" +
" keys text,\n" +
" keysLoad text,\n" +
" keysLoadFor text,\n" +
" PRIMARY KEY (executor_id, status, position, unique_position)" +
')', UTF8Type.instance));
}
@Override
public DataSet data()
{
AccordCommandStores commandStores = (AccordCommandStores) AccordService.instance().node().commandStores();
SimpleDataSet ds = new SimpleDataSet(metadata());
for (AccordExecutor executor : commandStores.executors())
{
int uniquePos = 0;
int executorId = executor.executorId();
AccordExecutor.TaskInfo prev = null;
for (AccordExecutor.TaskInfo info : executor.taskSnapshot())
{
if (prev != null && info.status() == prev.status() && info.position() == prev.position()) ++uniquePos;
else uniquePos = 0;
prev = info;
PreLoadContext preLoadContext = info.preLoadContext();
ds.row(executorId, info.status(), info.position(), uniquePos)
.column("description", info.describe())
.column("command_store_id", info.commandStoreId())
.column("txn_id", preLoadContext == null ? null : preLoadContext.primaryTxnId())
.column("txn_id_additional", preLoadContext == null ? null : preLoadContext.additionalTxnId())
.column("keys", preLoadContext == null ? null : preLoadContext.keys())
.column("keysLoad", preLoadContext == null ? null : preLoadContext.loadKeys())
.column("keysLoadFor", preLoadContext == null ? null : preLoadContext.loadKeysFor())
;
}
}
return ds;
}
}
// TODO (desired): human readable packed key tracker (but requires loading Txn, so might be preferable to only do conditionally)
public static final class CoordinationsTable extends AbstractVirtualTable
{
private CoordinationsTable()
{
super(parse(VIRTUAL_ACCORD_DEBUG, COORDINATIONS,
"Accord Coordination State",
"CREATE TABLE %s (\n" +
" txn_id int,\n" +
" kind text,\n" +
" coordination_id int,\n" +
" description text,\n" +
" nodes text,\n" +
" nodes_inflight text,\n" +
" nodes_contacted text,\n" +
" participants text,\n" +
" replies text,\n" +
" tracker text,\n" +
" PRIMARY KEY (txn_id, kind, coordination_id)" +
')', UTF8Type.instance));
}
@Override
public DataSet data()
{
Coordinations coordinations = AccordService.instance().node().coordinations();
SimpleDataSet ds = new SimpleDataSet(metadata());
for (Coordination c : coordinations)
{
ds.row(c.txnId(), c.kind().toString(), c.coordinationId())
.column("nodes", toStringOrNull(c.nodes()))
.column("nodes_inflight", toStringOrNull(c.inflight()))
.column("nodes_contacted", toStringOrNull(c.contacted()))
.column("description", c.describe())
.column("participants", toStringOrNull(c.scope()))
.column("replies", summarise(c.replies()))
.column("tracker", summarise(c.tracker()));
}
return ds;
}
private static String summarise(@Nullable SortedListMap<Node.Id, ?> replies)
{
if (replies == null)
return null;
return AbstractCoordination.summariseReplies(replies, 60);
}
private static String summarise(@Nullable AbstractTracker<?> tracker)
{
if (tracker == null)
return null;
return tracker.summariseTracker();
}
}
// TODO (desired): don't report null as "null"
public static final class CommandsForKeyTable extends AbstractVirtualTable implements AbstractVirtualTable.DataSet
{
@ -529,7 +651,7 @@ public class AccordDebugKeyspace extends VirtualKeyspace
private static void addRow(SimpleDataSet ds, int executorId, String scope, AccordCache.ImmutableStats stats)
{
ds.row(executorId, scope)
.column("queries", stats.queries)
.column("queries", stats.hits + stats.misses)
.column("hits", stats.hits)
.column("misses", stats.misses);
}
@ -1365,7 +1487,7 @@ public class AccordDebugKeyspace extends VirtualKeyspace
Node node = AccordService.instance().node();
if (Route.isFullRoute(route))
{
RecoverWithRoute.recover(node, node.someSequentialExecutor(), txnId, NotKnownToBeInvalid, (FullRoute<?>) route, null, LatentStoreSelector.standard(), (success, fail) -> {
PrepareRecovery.recover(node, node.someSequentialExecutor(), txnId, NotKnownToBeInvalid, (FullRoute<?>) route, null, LatentStoreSelector.standard(), (success, fail) -> {
if (fail != null) result.setFailure(fail);
else result.setSuccess(null);
}, node.agent().trace(txnId, RECOVER));

99
src/java/org/apache/cassandra/metrics/AccordCacheMetrics.java
View File

@ -18,35 +18,94 @@
package org.apache.cassandra.metrics;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.locks.Lock;
import java.util.function.Function;
import java.util.function.ToLongFunction;
import com.codahale.metrics.Histogram;
import com.codahale.metrics.Gauge;
import org.apache.cassandra.service.accord.AccordExecutor;
import org.apache.cassandra.service.accord.IAccordService;
import static org.apache.cassandra.metrics.CacheMetrics.TYPE_NAME;
import static org.apache.cassandra.metrics.AccordMetricUtils.fromAccordService;
import static org.apache.cassandra.metrics.CassandraMetricsRegistry.Metrics;
public class AccordCacheMetrics extends CacheAccessMetrics
public class AccordCacheMetrics
{
public static final String OBJECT_SIZE = "ObjectSize";
public static final String ACCORD_CACHE = "AccordCache";
public static final AccordCacheMetrics CommandsCacheMetrics = new AccordCacheMetrics("Commands");
public static final AccordCacheMetrics CommandsForKeyCacheMetrics = new AccordCacheMetrics("CommandsForKey");
public static final AccordCacheGlobalMetrics Global = new AccordCacheGlobalMetrics();
public final Histogram objectSize;
private final Map<String, CacheAccessMetrics> instanceMetrics = new ConcurrentHashMap<>(2);
private final String scope;
public AccordCacheMetrics(String scope)
// not sure why we create these wrapper objects that can only be instantiated once, but it's a pattern in this package so...
public static class AccordCacheGlobalMetrics
{
super(new DefaultNameFactory(TYPE_NAME, scope));
objectSize = Metrics.histogram(factory.createMetricName(OBJECT_SIZE), false);
this.scope = scope;
final Gauge<Long> usedBytes;
final Gauge<Long> unreferencedBytes;
public AccordCacheGlobalMetrics()
{
DefaultNameFactory factory = new DefaultNameFactory("AccordCache");
this.usedBytes = Metrics.gauge(factory.createMetricName("UsedBytes"), fromAccordService(sumExecutors(executor -> executor.cacheUnsafe().weightedSize()), 0L));
this.unreferencedBytes = Metrics.gauge(factory.createMetricName("UnreferencedBytes"), fromAccordService(sumExecutors(executor -> executor.cacheUnsafe().unreferencedBytes()), 0L));
}
private static Function<IAccordService, Long> sumExecutors(ToLongFunction<AccordExecutor> f)
{
return service -> {
long sum = 0;
for (AccordExecutor executor : service.executors())
sum += f.applyAsLong(executor);
return sum;
};
}
}
public CacheAccessMetrics forInstance(Class<?> klass)
public static class Shard
{
// cannot make Class<?> hashCode deterministic, as cannot rewrite - so cannot safely use as Map key if want deterministic simulation
// (or we need to create extra hoops to catch this specific case in method rewriting)
return instanceMetrics.computeIfAbsent(klass.getSimpleName(), k -> new CacheAccessMetrics(new DefaultNameFactory(TYPE_NAME, String.format("%s-%s", scope, k))));
public final ShardedHitRate.HitRateShard hitRate;
public final LogLinearHistogram objectSize;
public Shard(ShardedHitRate.HitRateShard hitRate, LogLinearHistogram objectSize)
{
this.hitRate = hitRate;
this.objectSize = objectSize;
}
}
public final ShardedHitRate hitRate = new ShardedHitRate();
public final ShardedHistogram objectSize;
public final Gauge<Long> hits;
public final Gauge<Long> misses;
public final Gauge<Long> requests;
public final Gauge<Double> requestRate1m;
public final Gauge<Double> requestRate5m;
public final Gauge<Double> requestRate15m;
public final Gauge<Double> hitRateAllTime;
public final Gauge<Double> hitRate1m;
public final Gauge<Double> hitRate5m;
public final Gauge<Double> hitRate15m;
private final String subTypeName;
public AccordCacheMetrics(String subTypeName)
{
DefaultNameFactory factory = new DefaultNameFactory("AccordCache", subTypeName);
this.objectSize = Metrics.shardedHistogram(factory.createMetricName("EntrySize"));
this.hits = Metrics.gauge(factory.createMetricName("Hits"), hitRate::totalHits);
this.misses = Metrics.gauge(factory.createMetricName("Misses"), hitRate::totalMisses);
this.requests = Metrics.gauge(factory.createMetricName("Requests"), hitRate::totalRequests);
this.requestRate1m = Metrics.gauge(factory.createMetricName("Requests"), () -> hitRate.requestsPerSecond(1));
this.requestRate5m = Metrics.gauge(factory.createMetricName("Requests"), () -> hitRate.requestsPerSecond(5));
this.requestRate15m = Metrics.gauge(factory.createMetricName(RatioGaugeSet.FIFTEEN_MINUTE + "RequestRate"), () -> hitRate.requestsPerSecond(15));
this.hitRate1m = Metrics.gauge(factory.createMetricName(RatioGaugeSet.ONE_MINUTE + "HitRate"), () -> hitRate.hitRate(1));
this.hitRate5m = Metrics.gauge(factory.createMetricName(RatioGaugeSet.FIVE_MINUTE + "HitRate"), () -> hitRate.hitRate(5));
this.hitRate15m = Metrics.gauge(factory.createMetricName(RatioGaugeSet.FIFTEEN_MINUTE + "HitRate"), () -> hitRate.hitRate(15));
this.hitRateAllTime = Metrics.gauge(factory.createMetricName("Misses"), hitRate::hitRateAllTime);
this.subTypeName = subTypeName;
}
public Shard newShard(Lock guardedBy)
{
return new Shard(hitRate.newShard(guardedBy), objectSize.newShard(guardedBy));
}
}

1
src/java/org/apache/cassandra/metrics/AccordClientRequestMetrics.java
View File

@ -72,5 +72,4 @@ public class AccordClientRequestMetrics extends ClientRequestMetrics
Metrics.remove(factory.createMetricName("Preempted"));
Metrics.remove(factory.createMetricName("TopologyMismatches"));
}
}

226
src/java/org/apache/cassandra/metrics/AccordMetrics.java → src/java/org/apache/cassandra/metrics/AccordCoordinatorMetrics.java
View File

@ -19,50 +19,34 @@
package org.apache.cassandra.metrics;
import java.lang.reflect.Field;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicLong;
import javax.annotation.Nullable;
import accord.api.CoordinatorEventListener;
import accord.api.LocalEventListener;
import accord.api.Result;
import accord.coordinate.ExecutePath;
import accord.impl.progresslog.DefaultProgressLog;
import accord.local.Command;
import accord.local.Node;
import accord.local.SafeCommandStore;
import accord.primitives.Ballot;
import accord.primitives.Deps;
import accord.primitives.PartialDeps;
import accord.primitives.Timestamp;
import accord.primitives.TxnId;
import com.codahale.metrics.Counting;
import com.codahale.metrics.Gauge;
import com.codahale.metrics.Histogram;
import com.codahale.metrics.Meter;
import com.codahale.metrics.Timer;
import org.apache.cassandra.service.accord.AccordService;
import org.apache.cassandra.service.accord.api.AccordTimeService;
import org.apache.cassandra.tracing.Tracing;
import static java.util.concurrent.TimeUnit.MICROSECONDS;
import static org.apache.cassandra.metrics.CassandraMetricsRegistry.Metrics;
public class AccordMetrics
public class AccordCoordinatorMetrics
{
public final static AccordMetrics readMetrics = new AccordMetrics("ro");
public final static AccordMetrics writeMetrics = new AccordMetrics("rw");
public static final String ACCORD_REPLICA = "AccordReplica";
public static final String REPLICA_STABLE_LATENCY = "StableLatency";
public static final String REPLICA_PREAPPLY_LATENCY = "PreApplyLatency";
public static final String REPLICA_APPLY_LATENCY = "ApplyLatency";
public static final String REPLICA_APPLY_DURATION = "ApplyDuration";
public static final String REPLICA_DEPENDENCIES = "Dependencies";
public static final String PROGRESS_LOG_SIZE = "ProgressLogSize";
public final static AccordCoordinatorMetrics readMetrics = new AccordCoordinatorMetrics("ro");
public final static AccordCoordinatorMetrics writeMetrics = new AccordCoordinatorMetrics("rw");
public static final String ACCORD_COORDINATOR = "AccordCoordinator";
public static final String COORDINATOR_EPOCHS = "Epochs";
public static final String COORDINATOR_KEYS = "Keys";
public static final String COORDINATOR_TABLES = "Tables";
public static final String COORDINATOR_DEPENDENCIES = "Dependencies";
public static final String COORDINATOR_PREACCEPT_LATENCY = "PreAcceptLatency";
public static final String COORDINATOR_EXECUTE_LATENCY = "ExecuteLatency";
@ -71,60 +55,47 @@ public class AccordMetrics
public static final String MEDIUM_PATHS = "MediumPaths";
public static final String SLOW_PATHS = "SlowPaths";
public static final String PREEMPTED = "Preempted";
public static final String REJECTED = "Rejected";
public static final String TIMEOUTS = "Timeouts";
public static final String INVALIDATIONS = "Invalidations";
public static final String RECOVERY_DELAY = "RecoveryDelay";
public static final String RECOVERY_TIME = "RecoveryTime";
public static final String FAST_PATH_TO_TOTAL = "FastPathToTotal";
/**
* The time between start on the coordinator and commit on this replica.
*/
public final Timer replicaStableLatency;
/**
* The time between start on the coordinator and arrival of the result on this replica.
*/
public final Timer replicaPreapplyLatency;
/**
* The time between start on the coordinator and application on this replica.
*/
public final Timer replicaApplyLatency;
/**
* TODO (expected): probably more interesting is latency from preapplied to apply;
* we already track local write latencies, whch this effectively duplicates (but including queueing latencies)
* Duration of applying changes.
*/
public final Timer replicaApplyDuration;
/**
* A histogram of the number of dependencies per transaction at this replica.
*/
public final Histogram replicaDependencies;
public final Gauge<Long> progressLogSize;
/**
* A histogram of the number of dependencies per transaction at this coordinator.
*/
public final Histogram coordinatorDependencies;
public final Histogram dependencies;
/**
* A histogram of the time to preaccept on this coordinator
*/
public final Histogram coordinatorPreacceptLatency;
public final Histogram preacceptLatency;
/**
* A histogram of the time to begin execution on this coordinator
*/
public final Histogram coordinatorExecuteLatency;
public final Histogram executeLatency;
/**
* A histogram of the time to complete execution on this coordinator
*/
public final Histogram coordinatorApplyLatency;
public final Histogram applyLatency;
/**
* The number of epochs used to coordinate the transaction
*/
public final Histogram epochs;
/**
* The number of keys involved in a transaction
*/
public final Histogram keys;
/**
* The number of tables involved in a transaction
*/
public final Histogram tables;
/**
* The number of fast path transactions executed on this coordinator.
@ -146,6 +117,11 @@ public class AccordMetrics
*/
public final Meter preempted;
/**
* The number of preempted transactions on this coordinator.
*/
public final Meter rejected;
/**
* The number of timed out transactions on this coordinator.
*/
@ -171,31 +147,22 @@ public class AccordMetrics
*/
public final RatioGaugeSet fastPathToTotal;
private AccordMetrics(String scope)
private AccordCoordinatorMetrics(String scope)
{
DefaultNameFactory replica = new DefaultNameFactory(ACCORD_REPLICA, scope);
replicaStableLatency = Metrics.timer(replica.createMetricName(REPLICA_STABLE_LATENCY));
replicaPreapplyLatency = Metrics.timer(replica.createMetricName(REPLICA_PREAPPLY_LATENCY));
replicaApplyLatency = Metrics.timer(replica.createMetricName(REPLICA_APPLY_LATENCY));
replicaApplyDuration = Metrics.timer(replica.createMetricName(REPLICA_APPLY_DURATION));
replicaDependencies = Metrics.histogram(replica.createMetricName(REPLICA_DEPENDENCIES), true);
progressLogSize = Metrics.gauge(replica.createMetricName(PROGRESS_LOG_SIZE), () -> {
AtomicLong i = new AtomicLong();
AccordService.instance().node().commandStores().forEachCommandStore(store -> {
i.addAndGet(((DefaultProgressLog)store.unsafeProgressLog()).size());
});
return i.get();
});
DefaultNameFactory coordinator = new DefaultNameFactory(ACCORD_COORDINATOR, scope);
coordinatorDependencies = Metrics.histogram(coordinator.createMetricName(COORDINATOR_DEPENDENCIES), true);
coordinatorPreacceptLatency = Metrics.histogram(coordinator.createMetricName(COORDINATOR_PREACCEPT_LATENCY), true);
coordinatorExecuteLatency = Metrics.histogram(coordinator.createMetricName(COORDINATOR_EXECUTE_LATENCY), true);
coordinatorApplyLatency = Metrics.histogram(coordinator.createMetricName(COORDINATOR_APPLY_LATENCY), true);
dependencies = Metrics.histogram(coordinator.createMetricName(COORDINATOR_DEPENDENCIES), true);
preacceptLatency = Metrics.histogram(coordinator.createMetricName(COORDINATOR_PREACCEPT_LATENCY), true);
executeLatency = Metrics.histogram(coordinator.createMetricName(COORDINATOR_EXECUTE_LATENCY), true);
applyLatency = Metrics.histogram(coordinator.createMetricName(COORDINATOR_APPLY_LATENCY), true);
epochs = Metrics.histogram(coordinator.createMetricName(COORDINATOR_EPOCHS), true);
keys = Metrics.histogram(coordinator.createMetricName(COORDINATOR_KEYS), true);
tables = Metrics.histogram(coordinator.createMetricName(COORDINATOR_TABLES), true);
fastPaths = Metrics.meter(coordinator.createMetricName(FAST_PATHS));
mediumPaths = Metrics.meter(coordinator.createMetricName(MEDIUM_PATHS));
slowPaths = Metrics.meter(coordinator.createMetricName(SLOW_PATHS));
preempted = Metrics.meter(coordinator.createMetricName(PREEMPTED));
rejected = Metrics.meter(coordinator.createMetricName(REJECTED));
timeouts = Metrics.meter(coordinator.createMetricName(TIMEOUTS));
invalidations = Metrics.meter(coordinator.createMetricName(INVALIDATIONS));
recoveryDelay = Metrics.timer(coordinator.createMetricName(RECOVERY_DELAY));
@ -207,7 +174,7 @@ public class AccordMetrics
public String toString()
{
StringBuilder builder = new StringBuilder();
builder.append("AccordMetrics [");
builder.append("AccordCoordinatorMetrics [");
try
{
@ -229,20 +196,11 @@ public class AccordMetrics
return builder.toString();
}
public static class Listener implements CoordinatorEventListener, LocalEventListener
public static class Listener implements CoordinatorEventListener
{
public final static Listener instance = new Listener(AccordMetrics.readMetrics, AccordMetrics.writeMetrics);
public static final Listener instance = new Listener();
private final AccordMetrics readMetrics;
private final AccordMetrics writeMetrics;
public Listener(AccordMetrics readMetrics, AccordMetrics writeMetrics)
{
this.readMetrics = readMetrics;
this.writeMetrics = writeMetrics;
}
private AccordMetrics forTransaction(TxnId txnId)
private AccordCoordinatorMetrics forTransaction(TxnId txnId)
{
if (txnId != null)
{
@ -254,57 +212,14 @@ public class AccordMetrics
return null;
}
@Override
public void onStable(SafeCommandStore safeStore, Command cmd)
{
Tracing.trace("Stable {} on {}", cmd.txnId(), safeStore.commandStore());
long now = AccordTimeService.nowMicros();
AccordMetrics metrics = forTransaction(cmd.txnId());
if (metrics != null)
{
long trxTimestamp = cmd.txnId().hlc();
metrics.replicaStableLatency.update(now - trxTimestamp, TimeUnit.MICROSECONDS);
}
}
@Override
public void onPreApplied(SafeCommandStore safeStore, Command cmd)
{
Tracing.trace("Preapplied {} on {}", cmd.txnId(), safeStore.commandStore());
long now = AccordTimeService.nowMicros();
AccordMetrics metrics = forTransaction(cmd.txnId());
if (metrics != null)
{
Timestamp trxTimestamp = cmd.txnId();
metrics.replicaPreapplyLatency.update(now - trxTimestamp.hlc(), TimeUnit.MICROSECONDS);
PartialDeps deps = cmd.partialDeps();
metrics.replicaDependencies.update(deps != null ? deps.txnIdCount() : 0);
}
}
@Override
public void onApplied(SafeCommandStore safeStore, Command cmd, long applyStartedAt)
{
Tracing.trace("Applied {} on {}", cmd.txnId(), safeStore.commandStore());
long now = AccordTimeService.nowMicros();
AccordMetrics metrics = forTransaction(cmd.txnId());
if (metrics != null)
{
Timestamp trxTimestamp = cmd.txnId();
metrics.replicaApplyLatency.update(now - trxTimestamp.hlc(), TimeUnit.MICROSECONDS);
if (applyStartedAt > 0)
metrics.replicaApplyDuration.update(now - applyStartedAt, TimeUnit.MICROSECONDS);
}
}
@Override
public void onPreAccepted(TxnId txnId)
{
AccordMetrics metrics = forTransaction(txnId);
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
{
long now = AccordTimeService.nowMicros();
metrics.coordinatorPreacceptLatency.update(Math.max(0, now - txnId.hlc()));
metrics.preacceptLatency.update(Math.max(0, now - txnId.hlc()));
}
}
@ -312,12 +227,12 @@ public class AccordMetrics
public void onExecuting(TxnId txnId, @Nullable Ballot ballot, Deps deps, @Nullable ExecutePath path)
{
Tracing.trace("{} agreed {}", path, txnId);
AccordMetrics metrics = forTransaction(txnId);
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
{
metrics.coordinatorDependencies.update(deps.txnIdCount());
metrics.dependencies.update(deps.txnIdCount());
long now = AccordTimeService.nowMicros();
metrics.coordinatorExecuteLatency.update(Math.max(0, now - txnId.hlc()));
metrics.executeLatency.update(Math.max(0, now - txnId.hlc()));
if (path != null)
{
switch (path)
@ -333,18 +248,18 @@ public class AccordMetrics
@Override
public void onExecuted(TxnId txnId, Ballot ballot)
{
AccordMetrics metrics = forTransaction(txnId);
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
{
long now = AccordTimeService.nowMicros();
metrics.coordinatorApplyLatency.update(Math.max(0, now - txnId.hlc()));
metrics.applyLatency.update(Math.max(0, now - txnId.hlc()));
}
}
@Override
public void onRecoveryStopped(Node node, TxnId txnId, Ballot ballot, Result result, Throwable failure)
{
AccordMetrics metrics = forTransaction(txnId);
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
{
long now = AccordTimeService.nowMicros();
@ -358,9 +273,42 @@ public class AccordMetrics
public void onInvalidated(TxnId txnId)
{
Tracing.trace("Invalidated {}", txnId);
AccordMetrics metrics = forTransaction(txnId);
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
metrics.invalidations.mark();
}
@Override
public void onTimeout(@Nullable TxnId txnId)
{
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
metrics.timeouts.mark();
}
@Override
public void onRejected(TxnId txnId)
{
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
metrics.rejected.mark();
}
@Override
public void onExhausted(@Nullable TxnId txnId)
{
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
metrics.timeouts.mark();
CoordinatorEventListener.super.onExhausted(txnId);
}
@Override
public void onPreempted(@Nullable TxnId txnId)
{
AccordCoordinatorMetrics metrics = forTransaction(txnId);
if (metrics != null)
metrics.preempted.mark();
}
}
}

63
src/java/org/apache/cassandra/metrics/AccordMetricUtils.java Normal file
View File

@ -0,0 +1,63 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.util.function.Function;
import accord.local.durability.DurabilityService;
import accord.topology.TopologyManager;
import com.codahale.metrics.Gauge;
import org.apache.cassandra.service.accord.AccordService;
import org.apache.cassandra.service.accord.IAccordService;
public class AccordMetricUtils
{
static <T> Gauge<T> fromAccordService(Function<IAccordService, T> ifSetup, T ifNotSetup)
{
return fromAccordService(ifSetup, Function.identity(), ifNotSetup);
}
static Gauge<Long> fromTopologyManager(Function<TopologyManager, Long> ifSetup)
{
return fromAccordService(service -> service.node().topology(), ifSetup, 0L);
}
static Gauge<Long> fromDurabilityService(Function<DurabilityService, Long> ifSetup)
{
return fromAccordService(service -> service.node().durability(), ifSetup, 0L);
}
static <I, T> Gauge<T> fromAccordService(Function<IAccordService, I> initialStep, Function<I, T> finalStep, T ifNotSetup)
{
if (AccordService.isSetup())
{
IAccordService service = AccordService.instance();
if (!service.isEnabled())
return () -> ifNotSetup;
I intermediate = initialStep.apply(service);
return () -> finalStep.apply(intermediate);
}
return () -> {
IAccordService service = AccordService.instance();
if (!service.isEnabled())
return ifNotSetup;
return finalStep.apply(initialStep.apply(service));
};
}
}

175
src/java/org/apache/cassandra/metrics/AccordReplicaMetrics.java Normal file
View File

@ -0,0 +1,175 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.lang.reflect.Field;
import java.util.concurrent.TimeUnit;
import accord.api.ReplicaEventListener;
import accord.local.Command;
import accord.local.SafeCommandStore;
import accord.primitives.PartialDeps;
import accord.primitives.Timestamp;
import accord.primitives.TxnId;
import com.codahale.metrics.Counting;
import com.codahale.metrics.Histogram;
import com.codahale.metrics.Timer;
import org.apache.cassandra.service.accord.api.AccordTimeService;
import org.apache.cassandra.tracing.Tracing;
import static org.apache.cassandra.metrics.CassandraMetricsRegistry.Metrics;
public class AccordReplicaMetrics
{
public final static AccordReplicaMetrics readMetrics = new AccordReplicaMetrics("ro");
public final static AccordReplicaMetrics writeMetrics = new AccordReplicaMetrics("rw");
public static final String ACCORD_REPLICA = "AccordReplica";
public static final String REPLICA_STABLE_LATENCY = "StableLatency";
public static final String REPLICA_PREAPPLY_LATENCY = "PreApplyLatency";
public static final String REPLICA_APPLY_LATENCY = "ApplyLatency";
public static final String REPLICA_APPLY_DURATION = "ApplyDuration";
public static final String REPLICA_DEPENDENCIES = "Dependencies";
/**
* The time between start on the coordinator and commit on this replica.
*/
public final Timer stableLatency;
/**
* The time between start on the coordinator and arrival of the result on this replica.
*/
public final Timer preapplyLatency;
/**
* The time between start on the coordinator and application on this replica.
*/
public final Timer applyLatency;
/**
* TODO (expected): probably more interesting is latency from preapplied to apply;
* we already track local write latencies, whch this effectively duplicates (but including queueing latencies)
* Duration of applying changes.
*/
public final Timer applyDuration;
/**
* A histogram of the number of dependencies per transaction at this replica.
*/
public final Histogram dependencies;
private AccordReplicaMetrics(String scope)
{
DefaultNameFactory replica = new DefaultNameFactory(ACCORD_REPLICA, scope);
stableLatency = Metrics.timer(replica.createMetricName(REPLICA_STABLE_LATENCY));
preapplyLatency = Metrics.timer(replica.createMetricName(REPLICA_PREAPPLY_LATENCY));
applyLatency = Metrics.timer(replica.createMetricName(REPLICA_APPLY_LATENCY));
applyDuration = Metrics.timer(replica.createMetricName(REPLICA_APPLY_DURATION));
dependencies = Metrics.histogram(replica.createMetricName(REPLICA_DEPENDENCIES), true);
}
@Override
public String toString()
{
StringBuilder builder = new StringBuilder();
builder.append("AccordReplicaMetrics [");
try
{
for (Field f : getClass().getDeclaredFields())
{
f.setAccessible(true);
if (Counting.class.isAssignableFrom(f.getType()))
{
Counting metric = (Counting) f.get(this);
builder.append(String.format("%s: count=%d, ", f.getName(), metric.getCount()));
}
}
}
catch (IllegalAccessException e)
{
throw new RuntimeException(e);
}
builder.append("]");
return builder.toString();
}
public static class Listener implements ReplicaEventListener
{
public static final Listener instance = new Listener();
private AccordReplicaMetrics forTransaction(TxnId txnId)
{
if (txnId != null)
{
if (txnId.isWrite())
return writeMetrics;
else if (txnId.isSomeRead())
return readMetrics;
}
return null;
}
@Override
public void onStable(SafeCommandStore safeStore, Command cmd)
{
Tracing.trace("Stable {} on {}", cmd.txnId(), safeStore.commandStore());
long now = AccordTimeService.nowMicros();
AccordReplicaMetrics metrics = forTransaction(cmd.txnId());
if (metrics != null)
{
long trxTimestamp = cmd.txnId().hlc();
metrics.stableLatency.update(now - trxTimestamp, TimeUnit.MICROSECONDS);
}
}
@Override
public void onPreApplied(SafeCommandStore safeStore, Command cmd)
{
Tracing.trace("Preapplied {} on {}", cmd.txnId(), safeStore.commandStore());
long now = AccordTimeService.nowMicros();
AccordReplicaMetrics metrics = forTransaction(cmd.txnId());
if (metrics != null)
{
Timestamp trxTimestamp = cmd.txnId();
metrics.preapplyLatency.update(now - trxTimestamp.hlc(), TimeUnit.MICROSECONDS);
PartialDeps deps = cmd.partialDeps();
metrics.dependencies.update(deps != null ? deps.txnIdCount() : 0);
}
}
@Override
public void onApplied(SafeCommandStore safeStore, Command cmd, long applyStartedAt)
{
Tracing.trace("Applied {} on {}", cmd.txnId(), safeStore.commandStore());
long now = AccordTimeService.nowMicros();
AccordReplicaMetrics metrics = forTransaction(cmd.txnId());
if (metrics != null)
{
Timestamp trxTimestamp = cmd.txnId();
metrics.applyLatency.update(now - trxTimestamp.hlc(), TimeUnit.MICROSECONDS);
if (applyStartedAt > 0)
metrics.applyDuration.update(now - applyStartedAt, TimeUnit.MICROSECONDS);
}
}
}
// to ensure initialised
public static void touch() {}
}

233
src/java/org/apache/cassandra/metrics/AccordSystemMetrics.java Normal file
View File

@ -0,0 +1,233 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.lang.reflect.Field;
import java.util.concurrent.TimeUnit;
import accord.impl.progresslog.DefaultProgressLog;
import accord.local.MaxDecidedRX;
import accord.local.RedundantBefore;
import accord.primitives.TxnId;
import accord.topology.TopologyManager;
import accord.utils.Invariants;
import com.codahale.metrics.Counting;
import com.codahale.metrics.Gauge;
import org.apache.cassandra.metrics.LogLinearHistogram.LogLinearSnapshot;
import org.apache.cassandra.service.accord.AccordService;
import org.apache.cassandra.service.accord.IAccordService;
import org.apache.cassandra.utils.Clock;
import static accord.local.RedundantStatus.Property.GC_BEFORE;
import static accord.local.RedundantStatus.Property.LOCALLY_APPLIED;
import static accord.local.RedundantStatus.Property.LOCALLY_DURABLE_TO_COMMAND_STORE;
import static accord.local.RedundantStatus.Property.LOCALLY_DURABLE_TO_DATA_STORE;
import static accord.local.RedundantStatus.Property.QUORUM_APPLIED;
import static accord.local.RedundantStatus.Property.SHARD_APPLIED;
import static org.apache.cassandra.metrics.AccordMetricUtils.fromDurabilityService;
import static org.apache.cassandra.metrics.AccordMetricUtils.fromTopologyManager;
import static org.apache.cassandra.metrics.CassandraMetricsRegistry.Metrics;
public class AccordSystemMetrics
{
public final static AccordSystemMetrics metrics = new AccordSystemMetrics();
private static final long REFRESH_RATE = TimeUnit.SECONDS.toNanos(30);
public static final String ACCORD_SYSTEM = "AccordSystem";
public static final String MIN_EPOCH = "MinEpoch";
public static final String MAX_EPOCH = "MaxEpoch";
public static final String PROGRESS_LOG_ACTIVE = "ProgressLogActive";
public static final String PROGRESS_LOG_SIZE = "ProgressLogSize";
public static final String PROGRESS_LOG_AGE = "ProgressLogAge";
public static final String DURABILITY_QUEUE_ACTIVE = "DurabilityQueueActive";
public static final String DURABILITY_QUEUE_PENDING = "DurabilityQueuePending";
public static final String SYNCPOINT_AGREED_LAG = "SyncPointAgreedLag";
public static final String LOCALLY_APPLIED_LAG = "LocallyAppliedLag";
public static final String LOCALLY_DURABLE_LAG = "LocallyDurableLag";
public static final String QUORUM_APPLIED_LAG = "QuorumAppliedLag";
public static final String SHARD_APPLIED_LAG = "ShardAppliedLag";
public static final String GC_LAG = "GCLag";
public final Gauge<Long> minEpoch;
public final Gauge<Long> maxEpoch;
public final Gauge<Long> progressLogActive;
public final Gauge<Long> durabilityQueueActive;
public final Gauge<Long> durabilityQueuePending;
public final OnDemandHistogram progressLogSize;
public final OnDemandHistogram progressLogAge;
public final OnDemandHistogram syncPointAgreedLag;
public final OnDemandHistogram locallyAppliedLag;
public final OnDemandHistogram locallyDurableLag;
public final OnDemandHistogram quorumAppliedLag;
public final OnDemandHistogram shardAppliedLag;
public final OnDemandHistogram gcLag;
private Snapshot snapshot;
static class Snapshot
{
final long at;
final long progressLogActive;
final LogLinearSnapshot progressLogSize;
final LogLinearSnapshot progressLogAge;
final LogLinearSnapshot syncPointAgreedLag;
final LogLinearSnapshot locallyAppliedLag;
final LogLinearSnapshot locallyDurableLag;
final LogLinearSnapshot quorumAppliedLag;
final LogLinearSnapshot shardAppliedLag;
final LogLinearSnapshot gcLag;
Snapshot()
{
this.at = Clock.Global.nanoTime();
progressLogActive = 0;
progressLogSize = progressLogAge = syncPointAgreedLag = locallyAppliedLag = locallyDurableLag = gcLag = quorumAppliedLag = shardAppliedLag = new LogLinearSnapshot(0);
}
Snapshot(SnapshotBuilder builder)
{
this.at = Clock.Global.nanoTime();
this.progressLogActive = builder.progressLogActive;
this.progressLogSize = builder.progressLogSize.destroyToSnapshot();
this.progressLogAge = builder.progressLogAge.destroyToSnapshot();
this.syncPointAgreedLag = builder.syncPointAgreedLag.destroyToSnapshot();
this.locallyAppliedLag = builder.locallyAppliedLag.destroyToSnapshot();
this.locallyDurableLag = builder.locallyDurableLag.destroyToSnapshot();
this.quorumAppliedLag = builder.quorumAppliedLag.destroyToSnapshot();
this.shardAppliedLag = builder.shardAppliedLag.destroyToSnapshot();
this.gcLag = builder.gcLag.destroyToSnapshot();
}
}
static class SnapshotBuilder
{
long progressLogActive;
final LogLinearHistogram progressLogSize = new LogLinearHistogram(100);
final LogLinearHistogram progressLogAge = new LogLinearHistogram(TimeUnit.HOURS.toSeconds(1L));
final LogLinearHistogram syncPointAgreedLag = new LogLinearHistogram(TimeUnit.HOURS.toSeconds(1L));
final LogLinearHistogram locallyAppliedLag = new LogLinearHistogram(TimeUnit.HOURS.toSeconds(1L));
final LogLinearHistogram locallyDurableLag = new LogLinearHistogram(TimeUnit.HOURS.toSeconds(1L));
final LogLinearHistogram quorumAppliedLag = new LogLinearHistogram(TimeUnit.HOURS.toSeconds(1L));
final LogLinearHistogram shardAppliedLag = new LogLinearHistogram(TimeUnit.HOURS.toSeconds(1L));
final LogLinearHistogram gcLag = new LogLinearHistogram(TimeUnit.HOURS.toSeconds(1L));
SnapshotBuilder()
{
}
}
private AccordSystemMetrics()
{
Invariants.expect(AccordService.isSetup());
DefaultNameFactory factory = new DefaultNameFactory(ACCORD_SYSTEM);
minEpoch = Metrics.gauge(factory.createMetricName(MIN_EPOCH), fromTopologyManager(TopologyManager::minEpoch));
maxEpoch = Metrics.gauge(factory.createMetricName(MAX_EPOCH), fromTopologyManager(TopologyManager::epoch));
durabilityQueueActive = Metrics.gauge(factory.createMetricName(DURABILITY_QUEUE_ACTIVE), fromDurabilityService(durability -> (long)durability.queue().activeCount()));
durabilityQueuePending = Metrics.gauge(factory.createMetricName(DURABILITY_QUEUE_PENDING), fromDurabilityService(durability -> (long)durability.queue().pendingCount()));
progressLogActive = Metrics.gauge(factory.createMetricName(PROGRESS_LOG_ACTIVE), fromDurabilityService(durability -> (long)durability.queue().activeCount()));
progressLogSize = Metrics.onDemandHistogram(factory.createMetricName(PROGRESS_LOG_SIZE), () -> maybeRefreshHistograms().progressLogSize);
progressLogAge = Metrics.onDemandHistogram(factory.createMetricName(PROGRESS_LOG_AGE), () -> maybeRefreshHistograms().progressLogAge);
syncPointAgreedLag = Metrics.onDemandHistogram(factory.createMetricName(SYNCPOINT_AGREED_LAG), () -> maybeRefreshHistograms().syncPointAgreedLag);
locallyAppliedLag = Metrics.onDemandHistogram(factory.createMetricName(LOCALLY_APPLIED_LAG), () -> maybeRefreshHistograms().locallyAppliedLag);
locallyDurableLag = Metrics.onDemandHistogram(factory.createMetricName(LOCALLY_DURABLE_LAG), () -> maybeRefreshHistograms().locallyDurableLag);
quorumAppliedLag = Metrics.onDemandHistogram(factory.createMetricName(QUORUM_APPLIED_LAG), () -> maybeRefreshHistograms().quorumAppliedLag);
shardAppliedLag = Metrics.onDemandHistogram(factory.createMetricName(SHARD_APPLIED_LAG), () -> maybeRefreshHistograms().shardAppliedLag);
gcLag = Metrics.onDemandHistogram(factory.createMetricName(GC_LAG), () -> maybeRefreshHistograms().gcLag);
}
private synchronized Snapshot maybeRefreshHistograms()
{
if (snapshot == null || Clock.Global.nanoTime() >= snapshot.at + REFRESH_RATE)
refreshHistograms();
return snapshot;
}
private synchronized void refreshHistograms()
{
if (!AccordService.isSetup())
{
snapshot = new Snapshot();
return;
}
IAccordService service = AccordService.instance();
if (!service.isEnabled())
{
snapshot = new Snapshot();
return;
}
int nowSeconds = (int) (Clock.Global.currentTimeMillis() / 1000);
SnapshotBuilder builder = new SnapshotBuilder();
service.node().commandStores().forEachCommandStore(commandStore -> {
DefaultProgressLog.ImmutableView view = ((DefaultProgressLog)commandStore.unsafeProgressLog()).immutableView();
builder.progressLogActive += view.activeCount();
builder.progressLogSize.increment(view.size());
while (view.advance())
builder.progressLogAge.increment(ageSeconds(nowSeconds, view.txnId()));
RedundantBefore redundantBefore = commandStore.unsafeGetRedundantBefore();
for (int i = 0 ; i < redundantBefore.size() ; ++i)
{
RedundantBefore.Bounds bounds = redundantBefore.valueAt(i);
builder.locallyAppliedLag.increment(ageSeconds(nowSeconds, bounds.maxBound(LOCALLY_APPLIED)));
builder.locallyDurableLag.increment(ageSeconds(nowSeconds, bounds.maxBoundBoth(LOCALLY_DURABLE_TO_DATA_STORE, LOCALLY_DURABLE_TO_COMMAND_STORE)));
builder.quorumAppliedLag.increment(ageSeconds(nowSeconds, bounds.maxBound(QUORUM_APPLIED)));
builder.shardAppliedLag.increment(ageSeconds(nowSeconds, bounds.maxBound(SHARD_APPLIED)));
builder.gcLag.increment(ageSeconds(nowSeconds, bounds.maxBound(GC_BEFORE)));
}
MaxDecidedRX maxDecidedRX = commandStore.unsafeGetMaxDecidedRX();
for (int i = 0 ; i < maxDecidedRX.size() ; ++i)
builder.syncPointAgreedLag.increment(ageSeconds(nowSeconds, maxDecidedRX.valueAt(i).any));
});
this.snapshot = new Snapshot(builder);
}
static int ageSeconds(int nowSeconds, TxnId txnId)
{
return Math.max(0, nowSeconds - (int)TimeUnit.MICROSECONDS.toSeconds(txnId.hlc()));
}
@Override
public String toString()
{
StringBuilder builder = new StringBuilder();
builder.append("AccordSystemMetrics [");
try
{
for (Field f : getClass().getDeclaredFields())
{
f.setAccessible(true);
if (Counting.class.isAssignableFrom(f.getType()))
{
Counting metric = (Counting) f.get(this);
builder.append(String.format("%s: count=%d, ", f.getName(), metric.getCount()));
}
}
}
catch (IllegalAccessException e)
{
throw new RuntimeException(e);
}
builder.append(']');
return builder.toString();
}
}

22
src/java/org/apache/cassandra/metrics/SnapshottingReservoir.java → src/java/org/apache/cassandra/metrics/CassandraHistogram.java
View File

@ -18,10 +18,24 @@
package org.apache.cassandra.metrics;
import com.codahale.metrics.Reservoir;
import com.codahale.metrics.Snapshot;
import com.codahale.metrics.Histogram;
public interface SnapshottingReservoir extends Reservoir
public class CassandraHistogram extends Histogram
{
Snapshot getPercentileSnapshot();
final CassandraReservoir reservoir;
public CassandraHistogram(CassandraReservoir reservoir)
{
super(reservoir);
this.reservoir = reservoir;
}
public CassandraReservoir.BucketStrategy bucketStrategy()
{
return reservoir.bucketStrategy();
}
public long[] bucketStarts(int length)
{
return reservoir.buckets(length);
}
}

76
src/java/org/apache/cassandra/metrics/CassandraMetricsRegistry.java
View File

@ -29,6 +29,7 @@ import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.TimeUnit;
import java.util.function.Consumer;
import java.util.function.Function;
import java.util.function.Supplier;
import java.util.function.UnaryOperator;
import java.util.stream.Stream;
import javax.annotation.Nullable;
@ -114,8 +115,10 @@ public class CassandraMetricsRegistry extends MetricRegistry
// for virtual tables.
metricGroups = ImmutableSet.<String>builder()
.add(AbstractMetrics.TYPE)
.add(AccordMetrics.ACCORD_COORDINATOR)
.add(AccordMetrics.ACCORD_REPLICA)
.add(AccordCoordinatorMetrics.ACCORD_COORDINATOR)
.add(AccordCacheMetrics.ACCORD_CACHE)
.add(AccordReplicaMetrics.ACCORD_REPLICA)
.add(AccordSystemMetrics.ACCORD_SYSTEM)
.add(BatchMetrics.TYPE_NAME)
.add(BufferPoolMetrics.TYPE_NAME)
.add(CIDRAuthorizerMetrics.TYPE_NAME)
@ -169,8 +172,8 @@ public class CassandraMetricsRegistry extends MetricRegistry
return Long.toString(((Counter) metric).getCount());
else if (metric instanceof Gauge)
return getGaugeValue((Gauge) metric);
else if (metric instanceof Histogram)
return Double.toString(((Histogram) metric).getSnapshot().getMedian());
else if (metric instanceof CassandraHistogram)
return Double.toString(((CassandraHistogram) metric).getSnapshot().getMedian());
else if (metric instanceof Meter)
return Long.toString(((Meter) metric).getCount());
else if (metric instanceof Timer)
@ -245,7 +248,7 @@ public class CassandraMetricsRegistry extends MetricRegistry
"All metrics with type \"Histogram\"",
new HistogramMetricRowWalker(),
Metrics.getMetrics(),
Histogram.class::isInstance,
CassandraHistogram.class::isInstance,
HistogramMetricRow::new))
.add(createSinglePartitionedValueFiltered(VIRTUAL_METRICS,
"type_meter",
@ -316,18 +319,28 @@ public class CassandraMetricsRegistry extends MetricRegistry
return meter;
}
public Histogram histogram(MetricName name, boolean considerZeroes)
public CassandraHistogram histogram(MetricName name, boolean considerZeroes)
{
return register(name, new ClearableHistogram(new DecayingEstimatedHistogramReservoir(considerZeroes)));
}
public Histogram histogram(MetricName name, MetricName alias, boolean considerZeroes)
public CassandraHistogram histogram(MetricName name, MetricName alias, boolean considerZeroes)
{
Histogram histogram = histogram(name, considerZeroes);
CassandraHistogram histogram = histogram(name, considerZeroes);
register(alias, histogram);
return histogram;
}
public ShardedHistogram shardedHistogram(MetricName name)
{
return register(name, new ShardedHistogram());
}
public OnDemandHistogram onDemandHistogram(MetricName name, Supplier<LogLinearHistogram.LogLinearSnapshot> snapshot)
{
return register(name, new OnDemandHistogram(snapshot));
}
public <T extends Gauge<?>> T gauge(MetricName name, T gauge)
{
return register(name, gauge);
@ -362,14 +375,14 @@ public class CassandraMetricsRegistry extends MetricRegistry
return timer;
}
public static SnapshottingReservoir createReservoir(TimeUnit durationUnit)
public static CassandraReservoir createReservoir(TimeUnit durationUnit)
{
SnapshottingReservoir reservoir;
CassandraReservoir reservoir;
if (durationUnit != TimeUnit.NANOSECONDS)
{
SnapshottingReservoir underlying = new DecayingEstimatedHistogramReservoir(DecayingEstimatedHistogramReservoir.DEFAULT_ZERO_CONSIDERATION,
DecayingEstimatedHistogramReservoir.LOW_BUCKET_COUNT,
DecayingEstimatedHistogramReservoir.DEFAULT_STRIPE_COUNT);
CassandraReservoir underlying = new DecayingEstimatedHistogramReservoir(DecayingEstimatedHistogramReservoir.DEFAULT_ZERO_CONSIDERATION,
DecayingEstimatedHistogramReservoir.LOW_BUCKET_COUNT,
DecayingEstimatedHistogramReservoir.DEFAULT_STRIPE_COUNT);
// fewer buckets should suffice if timer is not based on nanos
reservoir = new ScalingReservoir(underlying,
// timer update values in nanos.
@ -518,8 +531,10 @@ public class CassandraMetricsRegistry extends MetricRegistry
mbean = new JmxGauge((Gauge<?>) metric, name);
else if (metric instanceof Counter)
mbean = new JmxCounter((Counter) metric, name);
else if (metric instanceof CassandraHistogram)
mbean = new JmxHistogram((CassandraHistogram) metric, name);
else if (metric instanceof Histogram)
mbean = new JmxHistogram((Histogram) metric, name);
throw new UnsupportedOperationException("Must supply a CassandraHistogram");
else if (metric instanceof Timer)
mbean = new JmxTimer((Timer) metric, name, TimeUnit.SECONDS, DEFAULT_TIMER_UNIT);
else if (metric instanceof Metered)
@ -637,14 +652,20 @@ public class CassandraMetricsRegistry extends MetricRegistry
long[] values();
long[] getRecentValues();
String bucketsId();
long[] rawBuckets(int count);
long[] rawValues();
}
private static class JmxHistogram extends AbstractBean implements JmxHistogramMBean
{
private final Histogram metric;
final CassandraHistogram metric;
private long[] last = null;
private JmxHistogram(Histogram metric, ObjectName objectName)
private JmxHistogram(CassandraHistogram metric, ObjectName objectName)
{
super(objectName);
this.metric = metric;
@ -719,7 +740,10 @@ public class CassandraMetricsRegistry extends MetricRegistry
@Override
public long[] values()
{
return metric.getSnapshot().getValues();
long[] values = metric.getSnapshot().getValues();
if (metric.bucketStrategy() == CassandraReservoir.BucketStrategy.log_linear)
values = metric.bucketStrategy().translateTo(CassandraReservoir.BucketStrategy.exp_12_nozero, values);
return values;
}
/**
@ -738,6 +762,24 @@ public class CassandraMetricsRegistry extends MetricRegistry
last = now;
return delta;
}
@Override
public String bucketsId()
{
return metric.bucketStrategy().name();
}
@Override
public long[] rawBuckets(int count)
{
return metric.bucketStarts(count);
}
@Override
public long[] rawValues()
{
return metric.getSnapshot().getValues();
}
}
/**

121
src/java/org/apache/cassandra/metrics/CassandraReservoir.java Normal file
View File

@ -0,0 +1,121 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.util.Arrays;
import java.util.function.IntFunction;
import java.util.function.LongFunction;
import accord.utils.Invariants;
import com.codahale.metrics.Reservoir;
import com.codahale.metrics.Snapshot;
import org.apache.cassandra.utils.EstimatedHistogram;
public interface CassandraReservoir extends Reservoir
{
enum BucketStrategy
{
none(false, length -> { throw new UnsupportedOperationException(); }, scale -> { throw new UnsupportedOperationException(); }),
exp_12(true, length -> EstimatedHistogram.newOffsets(length, true), scale -> EstimatedHistogram.newOffsetsWithScale(scale, true)),
exp_12_nozero(true, length -> EstimatedHistogram.newOffsets(length, false), scale -> EstimatedHistogram.newOffsetsWithScale(scale, false)),
log_linear(false, LogLinearHistogram::bucketsWithLength, LogLinearHistogram::bucketsWithScale);
final boolean overflows;
final IntFunction<long[]> bucketsWithLength;
final LongFunction<long[]> bucketsWithScale;
volatile long[] cachedBuckets;
BucketStrategy(boolean overflows, IntFunction<long[]> bucketsWithLength, LongFunction<long[]> bucketsWithScale)
{
this.overflows = overflows;
this.bucketsWithLength = bucketsWithLength;
this.bucketsWithScale = bucketsWithScale;
}
private long[] sharedBucketsWithLength(int length)
{
long[] buckets = cachedBuckets;
if (buckets == null || buckets.length < length)
cachedBuckets = buckets = this.bucketsWithLength.apply(length);
return buckets;
}
private long[] sharedBucketsWithScale(long scale)
{
long[] buckets = cachedBuckets;
if (buckets == null || buckets.length == 0 || buckets[buckets.length - 1] < scale)
cachedBuckets = buckets = this.bucketsWithScale.apply(scale);
return buckets;
}
public long[] translateTo(BucketStrategy toStrategy, long[] srcValues)
{
if (srcValues.length == 0)
return srcValues;
long[] srcBuckets = sharedBucketsWithLength(srcValues.length + 1);
long[] trgBuckets = toStrategy.sharedBucketsWithScale(srcBuckets[srcValues.length]);
return translate(srcBuckets, srcValues, overflows ? srcValues.length - 1 : srcValues.length, trgBuckets);
}
public static long[] translate(long[] srcBuckets, long[] srcValues, int srcLength, long[] trgBuckets)
{
Invariants.requireArgument(srcLength < srcBuckets.length);
int trgLength = Arrays.binarySearch(trgBuckets, srcBuckets[srcLength]);
if (trgLength < 0) trgLength = -1 - trgLength;
else trgLength++;
long[] trgValues = new long[trgLength];
long srcStart = srcBuckets[0], srcRemainder = srcValues[0], srcEnd = srcBuckets[1];
int si = 0, ti = 0;
while (true)
{
long trgEnd = trgBuckets[ti + 1];
if (trgEnd <= srcStart)
{
if (++ti == trgValues.length) break;
continue;
}
long end = Math.min(trgEnd, srcEnd);
if (end == srcEnd)
{
trgValues[ti] += srcRemainder;
if (++si == srcLength) break;
srcStart = srcEnd;
srcEnd = srcBuckets[si + 1];
srcRemainder = srcValues[si];
}
else
{
double ratio = (end - srcStart) / (double)(srcEnd - srcStart);
long incr = Math.round(ratio * srcRemainder);
trgValues[ti] += incr;
srcRemainder -= incr;
srcStart = end;
}
}
return trgValues;
}
}
Snapshot getPercentileSnapshot();
long[] buckets(int length);
BucketStrategy bucketStrategy();
}

6
src/java/org/apache/cassandra/metrics/ClearableHistogram.java
View File

@ -23,12 +23,10 @@ import java.util.concurrent.atomic.LongAdder;
import com.google.common.annotations.VisibleForTesting;
import com.codahale.metrics.Histogram;
/**
* Adds ability to reset a histogram
*/
public class ClearableHistogram extends Histogram
public class ClearableHistogram extends CassandraHistogram
{
private final DecayingEstimatedHistogramReservoir reservoirRef;
@ -58,7 +56,7 @@ public class ClearableHistogram extends Histogram
// clearly if we start using an incompatible version of the metrics.
try
{
Field countField = Histogram.class.getDeclaredField("count");
Field countField = com.codahale.metrics.Histogram.class.getDeclaredField("count");
countField.setAccessible(true);
// in 3.1 the counter object is a LongAdderAdapter which is a package private interface
// from com.codahale.metrics. In 4.0, it is a java LongAdder so the code will be simpler.

6
src/java/org/apache/cassandra/metrics/ClientMetrics.java
View File

@ -31,9 +31,7 @@ import java.util.function.Predicate;
import com.google.common.annotations.VisibleForTesting;
import com.codahale.metrics.Gauge;
import com.codahale.metrics.Histogram;
import com.codahale.metrics.Meter;
import com.codahale.metrics.Reservoir;
import com.codahale.metrics.Timer;
import org.apache.cassandra.auth.AuthenticatedUser;
import org.apache.cassandra.auth.IAuthenticator;
@ -199,9 +197,9 @@ public final class ClientMetrics
registerGauge("ClientsByProtocolVersion", "clientsByProtocolVersion", this::recentClientStats);
registerGauge("RequestsSize", ClientResourceLimits::getCurrentGlobalUsage);
Reservoir ipUsageReservoir = ClientResourceLimits.ipUsageReservoir();
CassandraReservoir ipUsageReservoir = ClientResourceLimits.ipUsageReservoir();
Metrics.register(factory.createMetricName("RequestsSizeByIpDistribution"),
new Histogram(ipUsageReservoir)
new CassandraHistogram(ipUsageReservoir)
{
public long getCount()
{

16
src/java/org/apache/cassandra/metrics/DecayingEstimatedHistogramReservoir.java
View File

@ -84,7 +84,7 @@ import static org.apache.cassandra.config.CassandraRelevantProperties.DECAYING_E
*
* @see ExponentiallyDecayingReservoir
*/
public class DecayingEstimatedHistogramReservoir implements SnapshottingReservoir
public class DecayingEstimatedHistogramReservoir implements CassandraReservoir
{
private static final Logger logger = LoggerFactory.getLogger(DecayingEstimatedHistogramReservoir.class);
private static final NoSpamLogger noSpamLogger = NoSpamLogger.getLogger(logger, 5L, TimeUnit.MINUTES);
@ -343,6 +343,20 @@ public class DecayingEstimatedHistogramReservoir implements SnapshottingReservoi
return rescaleIfNeeded(clock.now());
}
@Override
public long[] buckets(int length)
{
if (length == bucketOffsets.length)
return bucketOffsets;
return EstimatedHistogram.newOffsets(length, bucketOffsets[0] == 0);
}
@Override
public BucketStrategy bucketStrategy()
{
return bucketOffsets[0] == 0 ? BucketStrategy.exp_12 : BucketStrategy.exp_12_nozero;
}
/**
* @return true if this histogram has overflowed -- that is, a value larger than our largest bucket could bound was added
*/

288
src/java/org/apache/cassandra/metrics/LogLinearHistogram.java Normal file
View File

@ -0,0 +1,288 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.io.OutputStream;
import java.io.OutputStreamWriter;
import java.io.PrintWriter;
import java.nio.charset.StandardCharsets;
import java.util.Arrays;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.primitives.Ints;
import accord.utils.Invariants;
import com.codahale.metrics.Snapshot;
/**
* A simple single-threaded histogram with log-linear buckets.
* This has approximately the same accuracy as the lg 1.2 growth of EstimatedHistogram, but is simpler and faster.
* This histogram also importantly supports decrements and replace (decrement/increment pairs), and auto grows if necessary,
* so no overflow and can be initialised to a small size to avoid wasting memory
*
* TODO (desired): improve performance and memory locality by using a small buffer for collecting updates with e.g. 4 bits per counter,
*/
public class LogLinearHistogram
{
private static final int MAX_INDEX = 247;
private long[] buckets;
long totalCount;
public LogLinearHistogram(long expectedMaxValue)
{
buckets = new long[buckets(expectedMaxValue)];
}
public void increment(long value)
{
int index = index(value);
buckets(index)[index]++;
++totalCount;
}
public void decrement(long value)
{
int index = index(value);
buckets(index)[index]--;
--totalCount;
}
public void replace(long decrement, long increment)
{
int decrementIndex = index(decrement);
int incrementIndex = index(increment);
long[] buckets = buckets(Math.max(decrementIndex, incrementIndex));
if (decrementIndex != incrementIndex)
{
--buckets[decrementIndex];
++buckets[incrementIndex];
}
}
public static long[] bucketsWithLength(int length)
{
Invariants.require(length <= MAX_INDEX + 1);
long[] buckets = new long[length];
for (int i = 0 ; i < length ; ++i)
buckets[i] = invertIndex(i);
return buckets;
}
public static long[] bucketsWithScale(long maxScale)
{
return bucketsWithLength(1 + index(maxScale));
}
private long[] buckets(int withIndexAtLeast)
{
if (buckets.length <= withIndexAtLeast)
{
Invariants.require(withIndexAtLeast <= MAX_INDEX);
buckets = Arrays.copyOf(buckets, (withIndexAtLeast | 0x3) + 1);
}
return buckets;
}
private static int buckets(long maxValue)
{
return 1 + (index(maxValue) | 0x3);
}
private static int index(long value)
{
if (value < 4)
return (int) value;
int log = 61 - Long.numberOfLeadingZeros(value);
int linear = (int) (value >>> log) & 0x3;
return (log + 1) * 4 + linear;
}
private static long invertIndex(int index)
{
if (index < 4)
return index;
int log = index / 4;
int linear = index & 0x3;
return (4L | linear) << (log - 1);
}
static
{
for (int i = 0 ; i < MAX_INDEX ; ++i)
{
Invariants.require(index(invertIndex(i)) == i);
Invariants.require(i == 0 || index(invertIndex(i) - 1) == i - 1);
}
Invariants.require(index(invertIndex(MAX_INDEX + 1) - 1) == MAX_INDEX);
}
public static class LogLinearSnapshot extends Snapshot
{
long totalCount;
long[] raw;
long[] cumulative;
public static LogLinearSnapshot emptyForMax(long maxValue)
{
return new LogLinearSnapshot(buckets(maxValue));
}
LogLinearSnapshot(int size)
{
this.raw = new long[size];
}
@VisibleForTesting
LogLinearSnapshot(long[] raw, long totalCount)
{
this.raw = raw;
this.totalCount = totalCount;
}
private long[] cumulative()
{
if (cumulative == null)
{
cumulative = new long[raw.length];
long sum = 0;
for (int i = 0 ; i < cumulative.length ; ++i)
cumulative[i] = sum += cumulative[i];
}
return cumulative;
}
private long get(long tot)
{
if (totalCount == 0)
return 0;
long[] cumulative = cumulative();
int i = Arrays.binarySearch(cumulative, tot);
if (i >= 0) while (i > 0 && cumulative[i-1] == tot) --i;
else i = Math.max(0, -2 - i);
long prevValue = invertIndex(i);
long prevCount = cumulative[i];
long nextValue = invertIndex(i + 1);
long nextCount = i + 1 == cumulative.length ? totalCount : cumulative[i + 1];
long gap = tot - prevCount;
// should we use double arithmetic here to avoid overflow?
return prevValue + ((nextValue - prevValue) * gap) / (nextCount - prevCount);
}
@Override
public double getValue(double quantile)
{
return get(Math.round(totalCount * quantile));
}
@Override
public int size()
{
return Ints.saturatedCast(totalCount);
}
@Override
public long getMax()
{
return get(totalCount);
}
@Override
public double getMean()
{
if (totalCount <= 1)
return 0.0D;
return get(totalCount / 2);
}
@Override
public long getMin()
{
return get(1);
}
/**
* Get the estimated standard deviation of the values added to this reservoir.
*
* As values are collected in variable sized buckets, the actual deviation may be more or less than the value
* returned.
*
* @return an estimate of the standard deviation
*/
@Override
public double getStdDev()
{
if (totalCount <= 1)
{
return 0.0D;
}
else
{
double mean = this.getMean();
double sum = 0.0D;
for(int i = 0; i < raw.length; ++i)
{
long value = invertIndex(i);
double diff = value - mean;
sum += diff * diff * raw[i];
}
return Math.sqrt(sum / (totalCount - 1));
}
}
@Override
public void dump(OutputStream output)
{
try (PrintWriter out = new PrintWriter(new OutputStreamWriter(output, StandardCharsets.UTF_8)))
{
for(int i = 0; i < raw.length; ++i)
out.printf("%d%n", raw[i]);
}
}
@Override
public long[] getValues()
{
return raw;
}
}
public LogLinearSnapshot destroyToSnapshot()
{
LogLinearSnapshot result = new LogLinearSnapshot(buckets, totalCount);
buckets = null;
this.totalCount = 0;
return result;
}
public void updateSnapshot(LogLinearSnapshot snapshot)
{
if (snapshot.raw.length < buckets.length)
snapshot.raw = Arrays.copyOf(snapshot.raw, buckets.length);
long[] raw = snapshot.raw;
for (int i = 0 ; i < raw.length ; ++i)
raw[i] = raw[i] + buckets[i];
snapshot.totalCount += totalCount;
snapshot.cumulative = null;
}
}

59
src/java/org/apache/cassandra/metrics/OnDemandHistogram.java Normal file
View File

@ -0,0 +1,59 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.util.function.Supplier;
import com.codahale.metrics.Snapshot;
import org.apache.cassandra.metrics.LogLinearHistogram.LogLinearSnapshot;
import static org.apache.cassandra.metrics.CassandraReservoir.BucketStrategy.log_linear;
public class OnDemandHistogram extends OverrideHistogram
{
final Supplier<LogLinearSnapshot> snapshot;
protected OnDemandHistogram(Supplier<LogLinearSnapshot> snapshot)
{
this.snapshot = snapshot;
}
@Override
public synchronized long getCount()
{
return snapshot.get().totalCount;
}
@Override
public Snapshot getSnapshot()
{
return snapshot.get();
}
@Override
public CassandraReservoir.BucketStrategy bucketStrategy()
{
return log_linear;
}
@Override
public long[] bucketStarts(int length)
{
return LogLinearHistogram.bucketsWithLength(length);
}
}

50
src/java/org/apache/cassandra/metrics/OverrideHistogram.java Normal file
View File

@ -0,0 +1,50 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import com.codahale.metrics.Snapshot;
import org.agrona.UnsafeAccess;
public abstract class OverrideHistogram extends CassandraHistogram
{
private static final CassandraReservoir NO_RESERVOIR = new CassandraReservoir() {
@Override public Snapshot getPercentileSnapshot() { return null; }
@Override public long[] buckets(int length) { return new long[0]; }
@Override public BucketStrategy bucketStrategy() { return BucketStrategy.none; }
@Override public int size() { return 0; }
@Override public void update(long value) {}
@Override public Snapshot getSnapshot() { return null; }
};
protected OverrideHistogram()
{
super(NO_RESERVOIR);
try
{
UnsafeAccess.UNSAFE.putObject(this, UnsafeAccess.UNSAFE.objectFieldOffset(CassandraHistogram.class.getDeclaredField("count")), null);
}
catch (Throwable t)
{
// best effort, don't worry if this fails
}
}
public abstract CassandraReservoir.BucketStrategy bucketStrategy();
public abstract long[] bucketStarts(int length);
}

29
src/java/org/apache/cassandra/metrics/ScalingReservoir.java
View File

@ -17,17 +17,19 @@
*/
package org.apache.cassandra.metrics;
import java.util.function.LongUnaryOperator;
import com.codahale.metrics.Snapshot;
/**
* A reservoir that scales the values before updating.
*/
public class ScalingReservoir implements SnapshottingReservoir
public class ScalingReservoir implements CassandraReservoir
{
private final SnapshottingReservoir delegate;
private final ScaleFunction scaleFunc;
private final CassandraReservoir delegate;
private final LongUnaryOperator scaleFunc;
public ScalingReservoir(SnapshottingReservoir reservoir, ScaleFunction scaleFunc)
public ScalingReservoir(CassandraReservoir reservoir, LongUnaryOperator scaleFunc)
{
this.delegate = reservoir;
this.scaleFunc = scaleFunc;
@ -42,7 +44,7 @@ public class ScalingReservoir implements SnapshottingReservoir
@Override
public void update(long value)
{
delegate.update(scaleFunc.apply(value));
delegate.update(scaleFunc.applyAsLong(value));
}
@Override
@ -57,14 +59,15 @@ public class ScalingReservoir implements SnapshottingReservoir
return delegate.getPercentileSnapshot();
}
/**
* Scale the input value.
*
* Not using {@code java.util.function.Function<Long, Long>} to avoid auto-boxing.
*/
@FunctionalInterface
public static interface ScaleFunction
@Override
public BucketStrategy bucketStrategy()
{
long apply(long value);
return delegate.bucketStrategy();
}
@Override
public long[] buckets(int length)
{
return delegate.buckets(length);
}
}

146
src/java/org/apache/cassandra/metrics/ShardedHistogram.java Normal file
View File

@ -0,0 +1,146 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.util.concurrent.CopyOnWriteArrayList;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Lock;
import com.codahale.metrics.Snapshot;
import org.apache.cassandra.metrics.LogLinearHistogram.LogLinearSnapshot;
import org.apache.cassandra.utils.Clock;
import static org.apache.cassandra.metrics.CassandraReservoir.BucketStrategy.log_linear;
public class ShardedHistogram extends OverrideHistogram
{
private static final long REFRESH_RATE = TimeUnit.SECONDS.toNanos(15);
static class HistogramShard
{
final Lock lock;
final LogLinearHistogram histogram;
HistogramShard(Lock lock, LogLinearHistogram histogram)
{
this.lock = lock;
this.histogram = histogram;
}
long total()
{
lock.lock();
try
{
return histogram.totalCount;
}
finally
{
lock.unlock();
}
}
public void updateSnapshot(LogLinearSnapshot snapshot)
{
lock.lock();
try
{
histogram.updateSnapshot(snapshot);
}
finally
{
lock.unlock();
}
}
}
final CopyOnWriteArrayList<HistogramShard> shards = new CopyOnWriteArrayList<>();
final long initialMaxValue;
public ShardedHistogram()
{
this(1 << 16);
}
@Override
public CassandraReservoir.BucketStrategy bucketStrategy()
{
return log_linear;
}
@Override
public long[] bucketStarts(int length)
{
return LogLinearHistogram.bucketsWithLength(length);
}
public ShardedHistogram(long initialMaxValue)
{
this.initialMaxValue = initialMaxValue;
}
public LogLinearHistogram newShard(Lock guardedBy)
{
HistogramShard shard = new HistogramShard(guardedBy, new LogLinearHistogram(initialMaxValue));
shards.add(shard);
return shard.histogram;
}
private LogLinearSnapshot snapshot;
private long snapshotAt;
public synchronized void refresh()
{
refresh(Clock.Global.nanoTime());
}
private synchronized LogLinearSnapshot refresh(long now)
{
LogLinearSnapshot snapshot = LogLinearSnapshot.emptyForMax(initialMaxValue);
for (HistogramShard shard : shards)
shard.updateSnapshot(snapshot);
this.snapshot = snapshot;
this.snapshotAt = now;
return snapshot;
}
private synchronized LogLinearSnapshot maybeRefresh()
{
if (shards.isEmpty())
return snapshot = new LogLinearSnapshot(0);
long now = Clock.Global.nanoTime();
if (snapshot != null && snapshotAt + REFRESH_RATE >= now)
return snapshot;
return refresh(now);
}
@Override
public synchronized long getCount()
{
return maybeRefresh().totalCount;
}
@Override
public Snapshot getSnapshot()
{
return maybeRefresh();
}
}

248
src/java/org/apache/cassandra/metrics/ShardedHitRate.java Normal file
View File

@ -0,0 +1,248 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import java.util.Arrays;
import java.util.concurrent.CopyOnWriteArrayList;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Lock;
import accord.utils.Invariants;
import org.apache.cassandra.utils.Clock;
public class ShardedHitRate
{
private static final long REFRESH_RATE = TimeUnit.SECONDS.toNanos(15L);
private static final int BUCKETS = 16;
private static final int BUCKET_MASK = 0xF;
private static final long CLOCK_SLICE_SHIFT = 10;
private static final long CLOCK_SLICE = 1 << CLOCK_SLICE_SHIFT;
private static final long CLOCK_DIVISOR = 60_000_000_000L >>> CLOCK_SLICE_SHIFT;
private static long now()
{
return Clock.Global.nanoTime() / CLOCK_DIVISOR;
}
private static int atToMinutes(long at)
{
return (int) (at >>> CLOCK_SLICE_SHIFT);
}
public static class HitRateShard
{
final Lock lock;
long lastAt;
final long[] hits = new long[BUCKETS];
final long[] misses = new long[BUCKETS];
long totalHits, totalMisses;
public HitRateShard(Lock lock)
{
this.lock = lock;
}
void tick(long at)
{
if (lastAt >= at)
return;
int count = Math.min(BUCKETS, atToMinutes(at - lastAt));
for (int i = 1 ; i <= count ; ++i)
{
int index = (atToMinutes(lastAt) + i) & BUCKET_MASK;
hits[index] = misses[index] = 0;
}
lastAt = at;
}
// assumes the lock is held
public void markHitExclusive()
{
markHitExclusive(now());
}
public void markMissExclusive()
{
markMissExclusive(now());
}
void markHitExclusive(long at)
{
tick(at);
hits[atToMinutes(at) & BUCKET_MASK]++;
++totalHits;
}
public void markMissExclusive(long at)
{
tick(at);
misses[atToMinutes(at) & BUCKET_MASK]++;
++totalMisses;
}
void updateSnapshot(ShardedHitRate.Snapshot snapshot, long at)
{
lock.lock();
try
{
if (at > lastAt)
tick(at);
int srcIndex = atToMinutes(lastAt) & BUCKET_MASK;
double split = (at % CLOCK_SLICE) / (double)CLOCK_SLICE;
{
snapshot.hits[0] += hits[srcIndex];
snapshot.misses[0] += misses[srcIndex];
}
for (int i = 1 ; i < BUCKETS - 1 ; ++i)
{
srcIndex = (srcIndex - 1) & BUCKET_MASK;
long h = hits[srcIndex], m = misses[srcIndex];
long hSplit = Math.round(h * split), mSplit = Math.round(m * split);
snapshot.hits[i] += hSplit;
snapshot.misses[i] += mSplit;
snapshot.hits[i - 1] += h - hSplit;
snapshot.misses[i - 1] += m - mSplit;
}
{
srcIndex = (srcIndex - 1) & BUCKET_MASK;
long h = hits[srcIndex], m = misses[srcIndex];
long hSplit = (long) (h * split), mSplit = (long) (m * split);
snapshot.hits[BUCKETS - 2] += h - hSplit;
snapshot.misses[BUCKETS - 2] += m - mSplit;
}
snapshot.totalHits += totalHits;
snapshot.totalMisses += totalMisses;
}
finally
{
lock.unlock();
}
}
}
static class Snapshot
{
final long[] hits = new long[BUCKETS - 1];
final long[] misses = new long[BUCKETS - 1];
long totalHits, totalMisses;
long lastUpdated = Clock.Global.nanoTime() - REFRESH_RATE;
double hitRate(int minutes)
{
Invariants.requireArgument(minutes <= BUCKETS);
long misses = 0, hits = 0;
for (int i = 0 ; i < minutes; ++i)
{
hits += this.hits[i];
misses += this.misses[i];
}
long total = hits + misses;
if (total == 0)
return 0.0D;
return hits / (double)(total);
}
long requests(int minutes)
{
Invariants.requireArgument(minutes <= BUCKETS);
long requests = 0;
for (int i = 0 ; i < minutes; ++i)
requests += hits[i] + misses[i];
return requests;
}
void reset()
{
Arrays.fill(hits, 0);
Arrays.fill(misses, 0);
totalMisses = totalHits = 0;
}
}
final CopyOnWriteArrayList<HitRateShard> shards = new CopyOnWriteArrayList<>();
final Snapshot snapshot = new Snapshot();
public ShardedHitRate()
{
}
public HitRateShard newShard(Lock guardedBy)
{
HitRateShard shard = new HitRateShard(guardedBy);
shards.add(shard);
return shard;
}
private void maybeRefresh()
{
if (Clock.Global.nanoTime() - snapshot.lastUpdated >= REFRESH_RATE)
refresh();
}
public synchronized void refresh()
{
long at = now();
snapshot.reset();
for (HitRateShard shard : shards)
shard.updateSnapshot(snapshot, at);
}
public synchronized long totalHits()
{
maybeRefresh();
return snapshot.totalHits;
}
public synchronized long totalMisses()
{
maybeRefresh();
return snapshot.totalMisses;
}
public synchronized long totalRequests()
{
maybeRefresh();
return snapshot.totalHits + snapshot.totalMisses;
}
public double hitRateAllTime()
{
maybeRefresh();
long totalRequests = snapshot.totalHits + snapshot.totalMisses;
if (totalRequests == 0)
return 0.0D;
return snapshot.totalHits / (double)(snapshot.totalHits + snapshot.totalMisses);
}
public double hitRate(int minutes)
{
maybeRefresh();
return snapshot.hitRate(minutes);
}
public double requestsPerSecond(int minutes)
{
maybeRefresh();
return snapshot.requests(minutes) / (minutes * 60d);
}
}

6
src/java/org/apache/cassandra/metrics/SnapshottingTimer.java
View File

@ -24,14 +24,14 @@ import com.codahale.metrics.Timer;
public class SnapshottingTimer extends Timer
{
private final SnapshottingReservoir reservoir;
private final CassandraReservoir reservoir;
public SnapshottingTimer(SnapshottingReservoir reservoir)
public SnapshottingTimer(CassandraReservoir reservoir)
{
this(reservoir, Clock.defaultClock());
}
public SnapshottingTimer(SnapshottingReservoir reservoir, Clock clock)
public SnapshottingTimer(CassandraReservoir reservoir, Clock clock)
{
super(reservoir, clock);
this.reservoir = reservoir;

5
src/java/org/apache/cassandra/service/RetryStrategy.java
View File

@ -247,6 +247,11 @@ public class RetryStrategy implements WaitStrategy
return units.convert(result, MICROSECONDS);
}
public long computeMaxWait(int attempt, TimeUnit units)
{
return units.convert(max.getMaxMicros(attempt), MICROSECONDS);
}
public static RetryStrategy parse(String spec, LatencySourceFactory latencies)
{
return parse(spec, latencies, null);

17
src/java/org/apache/cassandra/service/TimeoutStrategy.java
View File

@ -103,12 +103,14 @@ public class TimeoutStrategy implements WaitStrategy
public interface Wait
{
long getMicros(int attempts);
long getMaxMicros(int attempts);
class Constant implements Wait
{
final long micros;
public Constant(long micros) { this.micros = micros; }
@Override public long getMicros(int attempts) { return micros; }
@Override public long getMaxMicros(int attempts) { return micros; }
}
class Modifying implements Wait
@ -127,18 +129,26 @@ public class TimeoutStrategy implements WaitStrategy
{
return modifier.modify(supplier.getMicros(), attempts);
}
@Override
public long getMaxMicros(int attempts)
{
return modifier.modify(supplier.getMaxMicros(), attempts);
}
}
}
public interface LatencySupplier
{
long getMicros();
long getMaxMicros();
class Constant implements LatencySupplier
{
final long micros;
public Constant(long micros) {this.micros = micros; }
@Override public long getMicros() { return micros; }
@Override public long getMaxMicros() { return micros; }
}
class Percentile implements LatencySupplier
@ -152,11 +162,8 @@ public class TimeoutStrategy implements WaitStrategy
this.percentile = percentile;
}
@Override
public long getMicros()
{
return latencies.get(percentile);
}
@Override public long getMicros() { return latencies.get(percentile); }
@Override public long getMaxMicros() { return latencies.get(1.0); }
}
}

81
src/java/org/apache/cassandra/service/accord/AccordCache.java
View File

@ -59,7 +59,8 @@ import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.exceptions.UnknownTableException;
import org.apache.cassandra.io.util.DataInputBuffer;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import org.apache.cassandra.metrics.CacheAccessMetrics;
import org.apache.cassandra.metrics.LogLinearHistogram;
import org.apache.cassandra.metrics.ShardedHitRate;
import org.apache.cassandra.service.accord.AccordCacheEntry.LoadExecutor;
import org.apache.cassandra.service.accord.AccordCacheEntry.Status;
import org.apache.cassandra.service.accord.events.CacheEvents;
@ -73,8 +74,6 @@ import static accord.utils.Invariants.require;
import static org.apache.cassandra.service.accord.AccordCacheEntry.Status.EVICTED;
import static org.apache.cassandra.service.accord.AccordCacheEntry.Status.LOADED;
import static org.apache.cassandra.service.accord.AccordCacheEntry.Status.MODIFIED;
import static org.apache.cassandra.service.accord.AccordCacheEntry.Status.SAVING;
import static org.apache.cassandra.service.accord.AccordCacheEntry.Status.WAITING_TO_SAVE;
/**
* Cache for AccordCommand and AccordCommandsForKey, available memory is shared between the two object types.
@ -122,20 +121,17 @@ public class AccordCache implements CacheSize
static class Stats
{
long queries;
long hits;
long misses;
}
public static final class ImmutableStats
{
public final long queries;
public final long hits;
public final long misses;
public ImmutableStats(Stats stats)
{
queries = stats.queries;
hits = stats.hits;
misses = stats.misses;
}
@ -146,22 +142,17 @@ public class AccordCache implements CacheSize
private final IntrusiveLinkedList<AccordCacheEntry<?,?>> evictQueue = new IntrusiveLinkedList<>();
private final IntrusiveLinkedList<AccordCacheEntry<?,?>> noEvictQueue = new IntrusiveLinkedList<>();
private int unreferencedBytes;
private long unreferencedBytes;
private int unreferenced;
private long maxSizeInBytes;
private long bytesCached;
private int noEvictGeneration;
private boolean shrinkingOn = true;
@VisibleForTesting
final AccordCacheMetrics metrics;
final Stats stats = new Stats();
public AccordCache(AccordCacheEntry.SaveExecutor saveExecutor, long maxSizeInBytes, AccordCacheMetrics metrics)
public AccordCache(AccordCacheEntry.SaveExecutor saveExecutor, long maxSizeInBytes)
{
this.saveExecutor = saveExecutor;
this.maxSizeInBytes = maxSizeInBytes;
this.metrics = metrics;
}
@Override
@ -299,6 +290,7 @@ public class AccordCache implements CacheSize
Type<?, ?, ?>.Instance owner = node.owner;
Type<?, ?, ?> parent = owner.parent();
parent.bytesCached -= node.sizeOnHeap;
parent.objectSize.decrement(node.sizeOnHeap);
--parent.size;
// TODO (expected): use listeners
@ -348,14 +340,9 @@ public class AccordCache implements CacheSize
safeRef.global().owner.release(safeRef, owner);
}
public ImmutableStats stats()
public <K, V, S extends AccordSafeState<K, V>> Type<K, V, S> newType(Class<K> keyClass, Adapter<K, V, S> adapter, AccordCacheMetrics.Shard metrics)
{
return new ImmutableStats(stats);
}
public <K, V, S extends AccordSafeState<K, V>> Type<K, V, S> newType(Class<K> keyClass, Adapter<K, V, S> adapter)
{
Type<K, V, S> instance = new Type<>(keyClass, adapter);
Type<K, V, S> instance = new Type<>(keyClass, adapter, metrics);
types.add(instance);
return instance;
}
@ -367,9 +354,10 @@ public class AccordCache implements CacheSize
Function<V, V> quickShrink,
TriFunction<AccordCommandStore, K, V, Boolean> validateFunction,
ToLongFunction<V> heapEstimator,
Function<AccordCacheEntry<K, V>, S> safeRefFactory)
Function<AccordCacheEntry<K, V>, S> safeRefFactory,
AccordCacheMetrics.Shard metrics)
{
return newType(keyClass, loadFunction, saveFunction, quickShrink, (i, j) -> j, (c, i, j) -> (V)j, validateFunction, heapEstimator, i -> 0, safeRefFactory);
return newType(keyClass, loadFunction, saveFunction, quickShrink, (i, j) -> j, (c, i, j) -> (V)j, validateFunction, heapEstimator, i -> 0, safeRefFactory, metrics);
}
public <K, V, S extends AccordSafeState<K, V>> Type<K, V, S> newType(
@ -382,12 +370,14 @@ public class AccordCache implements CacheSize
TriFunction<AccordCommandStore, K, V, Boolean> validateFunction,
ToLongFunction<V> heapEstimator,
ToLongFunction<Object> shrunkHeapEstimator,
Function<AccordCacheEntry<K, V>, S> safeRefFactory)
Function<AccordCacheEntry<K, V>, S> safeRefFactory,
AccordCacheMetrics.Shard metrics)
{
return newType(keyClass, new FunctionalAdapter<>(loadFunction, saveFunction, quickShrink,
fullShrink, inflate,
validateFunction, heapEstimator, shrunkHeapEstimator,
safeRefFactory, AccordCacheEntry::createReadyToLoad));
safeRefFactory, AccordCacheEntry::createReadyToLoad),
metrics);
}
public Collection<Type<?, ? ,? >> types()
@ -445,7 +435,6 @@ public class AccordCache implements CacheSize
private AccordCacheEntry<K, V> acquire(K key, boolean onlyIfLoaded)
{
incrementCacheQueries();
@SuppressWarnings("unchecked")
AccordCacheEntry<K, V> node = cache.get(key);
return node == null
@ -691,25 +680,24 @@ public class AccordCache implements CacheSize
private int size;
@VisibleForTesting
final CacheAccessMetrics typeMetrics;
final LogLinearHistogram objectSize;
final ShardedHitRate.HitRateShard hitRate;
private final Stats stats = new Stats();
private List<Listener<K, V>> typeListeners = null;
public Type(
Class<K> keyClass,
Adapter<K, V, S> adapter)
public Type(Class<K> keyClass, Adapter<K, V, S> adapter, AccordCacheMetrics.Shard metrics)
{
this.keyClass = keyClass;
this.adapter = adapter;
this.typeMetrics = metrics.forInstance(keyClass);
this.objectSize = metrics.objectSize;
this.hitRate = metrics.hitRate;
}
void updateSize(long newSize, long delta, boolean isUnreferenced, boolean updateHistogram)
{
// TODO (expected): deprecate this in favour of a histogram snapshot of any point in time
bytesCached += delta;
AccordCache.this.bytesCached += delta;
if (updateHistogram) metrics.objectSize.update(newSize);
if (updateHistogram) objectSize.replace(newSize - delta, newSize);
if (isUnreferenced) AccordCache.this.unreferencedBytes += delta;
}
@ -719,28 +707,16 @@ public class AccordCache implements CacheSize
return new Instance(commandStore);
}
private void incrementCacheQueries()
{
typeMetrics.requests.mark();
metrics.requests.mark();
stats.queries++;
AccordCache.this.stats.queries++;
}
private void incrementCacheHits()
{
typeMetrics.hits.mark();
metrics.hits.mark();
hitRate.markHitExclusive();
stats.hits++;
AccordCache.this.stats.hits++;
}
private void incrementCacheMisses()
{
typeMetrics.misses.mark();
metrics.misses.mark();
hitRate.markMissExclusive();
stats.misses++;
AccordCache.this.stats.misses++;
}
final AccordCache parent()
@ -758,11 +734,6 @@ public class AccordCache implements CacheSize
return new ImmutableStats(stats);
}
public Stats globalStats()
{
return AccordCache.this.stats;
}
@VisibleForTesting
public void unsafeSetLoadFunction(BiFunction<AccordCommandStore, K, V> loadFunction)
{
@ -896,7 +867,7 @@ public class AccordCache implements CacheSize
}
@VisibleForTesting
int unreferencedBytes()
public long unreferencedBytes()
{
return unreferencedBytes;
}
@ -957,7 +928,6 @@ public class AccordCache implements CacheSize
event.instanceAllocated = type.weightedSize();
AccordCache.Stats stats = type.stats();
event.instanceStatsQueries = stats.queries;
event.instanceStatsHits = stats.hits;
event.instanceStatsMisses = stats.misses;
@ -967,11 +937,6 @@ public class AccordCache implements CacheSize
event.globalCapacity = type.capacity();
event.globalAllocated = type.globalAllocated();
stats = type.globalStats();
event.globalStatsQueries = stats.queries;
event.globalStatsHits = stats.hits;
event.globalStatsMisses = stats.misses;
event.update();
}

1
src/java/org/apache/cassandra/service/accord/AccordCacheEntry.java
View File

@ -258,6 +258,7 @@ public class AccordCacheEntry<K, V> extends IntrusiveLinkedListNode
// TODO (expected): we aren't weighing the keys
sizeOnHeap = Ints.saturatedCast(EMPTY_SIZE);
parent.updateSize(sizeOnHeap, sizeOnHeap, false, false);
parent.objectSize.increment(EMPTY_SIZE);
}
@Override

2
src/java/org/apache/cassandra/service/accord/AccordCommandStore.java
View File

@ -190,7 +190,7 @@ public class AccordCommandStore extends CommandStore
this.caches = new ExclusiveCaches(sharedExecutor.lock, exclusive.global, commands, commandsForKey);
}
this.exclusiveExecutor = sharedExecutor.executor();
this.exclusiveExecutor = sharedExecutor.executor(id);
this.commandsForRanges = new CommandsForRanges.Manager(this);
maybeLoadRedundantBefore(journal.loadRedundantBefore(id()));

16
src/java/org/apache/cassandra/service/accord/AccordCommandStores.java
View File

@ -37,8 +37,6 @@ import accord.utils.RandomSource;
import org.apache.cassandra.cache.CacheSize;
import org.apache.cassandra.config.AccordSpec.QueueShardModel;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import org.apache.cassandra.metrics.CacheSizeMetrics;
import org.apache.cassandra.schema.TableId;
import org.apache.cassandra.service.accord.AccordExecutor.AccordExecutorFactory;
import org.apache.cassandra.service.accord.api.TokenKey;
@ -52,9 +50,6 @@ import static org.apache.cassandra.service.accord.AccordExecutor.constant;
public class AccordCommandStores extends CommandStores implements CacheSize
{
public static final String ACCORD_STATE_CACHE = "AccordStateCache";
private final CacheSizeMetrics cacheSizeMetrics;
private final AccordExecutor[] executors;
private final int mask;
@ -69,7 +64,6 @@ public class AccordCommandStores extends CommandStores implements CacheSize
super(node, agent, store, random, journal, shardDistributor, progressLogFactory, listenerFactory,
AccordCommandStore.factory(id -> executors[id % executors.length]));
this.executors = executors;
this.cacheSizeMetrics = new CacheSizeMetrics(ACCORD_STATE_CACHE, this);
this.mask = Integer.highestOneBit(executors.length) - 1;
cacheSize = DatabaseDescriptor.getAccordCacheSizeInMiB() << 20;
workingSetSize = DatabaseDescriptor.getAccordWorkingSetSizeInMiB() << 20;
@ -99,7 +93,6 @@ public class AccordCommandStores extends CommandStores implements CacheSize
for (int id = 0; id < executors.length; id++)
{
AccordCacheMetrics metrics = new AccordCacheMetrics(ACCORD_STATE_CACHE);
QueueShardModel shardModel = DatabaseDescriptor.getAccordQueueShardModel();
String baseName = AccordExecutor.class.getSimpleName() + '[' + id;
int threads = Math.min(maxThreads, Math.max(DatabaseDescriptor.getAccordConcurrentOps() / getAccordQueueShardCount(), 1));
@ -107,10 +100,10 @@ public class AccordCommandStores extends CommandStores implements CacheSize
{
case THREAD_PER_SHARD:
case THREAD_PER_SHARD_SYNC_QUEUE:
executors[id] = factory.get(id, shardModel == THREAD_PER_SHARD ? RUN_WITHOUT_LOCK : RUN_WITH_LOCK, 1, constant(baseName + ']'), metrics, agent);
executors[id] = factory.get(id, shardModel == THREAD_PER_SHARD ? RUN_WITHOUT_LOCK : RUN_WITH_LOCK, 1, constant(baseName + ']'), agent);
break;
case THREAD_POOL_PER_SHARD:
executors[id] = factory.get(id, RUN_WITHOUT_LOCK, threads, num -> baseName + ',' + num + ']', metrics, agent);
executors[id] = factory.get(id, RUN_WITHOUT_LOCK, threads, num -> baseName + ',' + num + ']', agent);
break;
}
}
@ -172,6 +165,11 @@ public class AccordCommandStores extends CommandStores implements CacheSize
refreshCapacities();
}
public synchronized void setShrinking(boolean on)
{
shrinkingOn = on;
}
@Override
public long capacity()
{

9
src/java/org/apache/cassandra/service/accord/AccordConfigurationService.java
View File

@ -120,6 +120,15 @@ public class AccordConfigurationService extends AbstractConfigurationService<Acc
{
return localSyncNotified;
}
public void addDebugString(StringBuilder sb)
{
sb.append(" syncStatus ")
.append(syncStatus)
.append(" localSyncNotified ")
.append(localSyncNotified);
super.addDebugString(sb);
}
}
static class EpochHistory extends AbstractConfigurationService.AbstractEpochHistory<EpochState>

138
src/java/org/apache/cassandra/service/accord/AccordExecutor.java
View File

@ -43,6 +43,7 @@ import accord.api.Agent;
import accord.api.AsyncExecutor;
import accord.api.RoutingKey;
import accord.local.Command;
import accord.local.PreLoadContext;
import accord.local.SequentialAsyncExecutor;
import accord.local.cfk.CommandsForKey;
import accord.primitives.TxnId;
@ -88,13 +89,14 @@ import static org.apache.cassandra.service.accord.AccordTask.State.WAITING_TO_RU
/**
* NOTE: We assume that NO BLOCKING TASKS are submitted to this executor AND WAITED ON by another task executing on this executor.
* (as we do not immediately schedule additional threads for submitted tasks, but schedule new threads only if necessary when the submitting execution completes)
*/
public abstract class AccordExecutor implements CacheSize, LoadExecutor<AccordTask<?>, Boolean>, SaveExecutor, Shutdownable, AsyncExecutor
{
private static final Logger logger = LoggerFactory.getLogger(AccordExecutor.class);
public interface AccordExecutorFactory
{
AccordExecutor get(int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent);
AccordExecutor get(int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent);
}
public enum Mode { RUN_WITH_LOCK, RUN_WITHOUT_LOCK }
@ -190,19 +192,19 @@ public abstract class AccordExecutor implements CacheSize, LoadExecutor<AccordTa
int tasks;
int runningThreads;
AccordExecutor(Lock lock, int executorId, AccordCacheMetrics metrics, Agent agent)
AccordExecutor(Lock lock, int executorId, Agent agent)
{
this.lock = lock;
this.executorId = executorId;
this.cache = new AccordCache(this, 0, metrics);
this.cache = new AccordCache(this, 0);
this.agent = agent;
final AccordCache.Type<TxnId, Command, AccordSafeCommand> commands;
final AccordCache.Type<RoutingKey, CommandsForKey, AccordSafeCommandsForKey> commandsForKey;
commands = cache.newType(TxnId.class, COMMAND_ADAPTER);
commands = cache.newType(TxnId.class, COMMAND_ADAPTER, AccordCacheMetrics.CommandsCacheMetrics.newShard(lock));
registerJfrListener(executorId, commands, "Command");
commandsForKey = cache.newType(RoutingKey.class, CFK_ADAPTER);
commandsForKey = cache.newType(RoutingKey.class, CFK_ADAPTER, AccordCacheMetrics.CommandsForKeyCacheMetrics.newShard(lock));
registerJfrListener(executorId, commandsForKey, "CommandsForKey");
this.caches = new ExclusiveGlobalCaches(this, cache, commands, commandsForKey);
@ -511,6 +513,11 @@ public abstract class AccordExecutor implements CacheSize, LoadExecutor<AccordTa
return new SequentialExecutor();
}
public SequentialExecutor executor(int commandStoreId)
{
return new SequentialExecutor(commandStoreId);
}
public SequentialAsyncExecutor newSequentialExecutor()
{
return new SequentialExecutor();
@ -962,14 +969,21 @@ public abstract class AccordExecutor implements CacheSize, LoadExecutor<AccordTa
private static final AtomicReferenceFieldUpdater<SequentialExecutor, Thread> ownerUpdater = AtomicReferenceFieldUpdater.newUpdater(SequentialExecutor.class, Thread.class, "owner");
public class SequentialExecutor extends TaskQueue<Task> implements SequentialAsyncExecutor
{
final int commandStoreId;
final SequentialQueueTask selfTask;
private Task task;
private volatile Thread owner, waiting;
private boolean running;
SequentialExecutor()
{
this(-1);
}
SequentialExecutor(int commandStoreId)
{
super(WAITING_TO_RUN);
this.commandStoreId = commandStoreId;
this.selfTask = new SequentialQueueTask(this);
}
@ -1180,6 +1194,11 @@ public abstract class AccordExecutor implements CacheSize, LoadExecutor<AccordTa
return peekNode();
}
protected T get(int index)
{
return super.get(index);
}
protected void remove(T remove)
{
super.remove(remove);
@ -1600,4 +1619,113 @@ public abstract class AccordExecutor implements CacheSize, LoadExecutor<AccordTa
return this;
}
}
public static class TaskInfo implements Comparable<TaskInfo>
{
public enum Status { WAITING_TO_LOAD, SCANNING_RANGES, LOADING, WAITING_TO_RUN, RUNNING }
final Status status;
final int commandStoreId;
final Task task;
public TaskInfo(Status status, int commandStoreId, Task task)
{
this.status = status;
this.commandStoreId = commandStoreId;
this.task = task;
}
public Status status()
{
return status;
}
public Integer commandStoreId()
{
return commandStoreId >= 0 ? commandStoreId : null;
}
public int position()
{
return task.queuePosition;
}
public @Nullable String describe()
{
if (task instanceof AccordTask)
return ((AccordTask<?>) task).preLoadContext().reason();
if (task instanceof DebuggableTask)
return ((DebuggableTask) task).description();
return null;
}
public @Nullable PreLoadContext preLoadContext()
{
if (task instanceof AccordTask)
return ((AccordTask<?>) task).preLoadContext();
return null;
}
@Override
public int compareTo(TaskInfo that)
{
int c = this.status.compareTo(that.status);
if (c == 0) c = this.position() - that.position();
return c;
}
}
public List<TaskInfo> taskSnapshot()
{
List<TaskInfo> result = new ArrayList<>();
lock.lock();
try
{
addToSnapshot(result, waitingToLoad, TaskInfo.Status.WAITING_TO_LOAD, TaskInfo.Status.WAITING_TO_LOAD);
addToSnapshot(result, waitingToLoadRangeTxns, TaskInfo.Status.WAITING_TO_LOAD, TaskInfo.Status.WAITING_TO_LOAD);
addToSnapshot(result, scanningRanges, TaskInfo.Status.SCANNING_RANGES, TaskInfo.Status.SCANNING_RANGES);
addToSnapshot(result, loading, TaskInfo.Status.LOADING, TaskInfo.Status.LOADING);
addToSnapshot(result, waitingToRun, TaskInfo.Status.WAITING_TO_RUN, TaskInfo.Status.WAITING_TO_RUN);
addToSnapshot(result, running, TaskInfo.Status.RUNNING, TaskInfo.Status.WAITING_TO_RUN);
}
finally
{
lock.unlock();
}
result.sort(TaskInfo::compareTo);
return result;
}
private static List<Task> toSimpleSnapshotList(TaskQueue<?> queue)
{
List<Task> list = new ArrayList<>();
for (int i = 0 ; i < queue.size() ; i++)
list.add(queue.get(i));
return list;
}
private static void addToSnapshot(List<TaskInfo> snapshot, TaskQueue<?> queue, TaskInfo.Status ifCurrent, TaskInfo.Status ifQueued)
{
for (int i = 0 ; i < queue.size() ; ++i)
{
Task t = queue.get(i);
if (t instanceof SequentialQueueTask)
{
SequentialExecutor q = ((SequentialQueueTask) t).queue;
snapshot.add(new TaskInfo(ifCurrent, q.commandStoreId, q.task));
for (int j = 0 ; j < q.size() ; ++j)
snapshot.add(new TaskInfo(ifQueued, q.commandStoreId, q.get(j)));
}
else
{
int commmandStoreId = t instanceof AccordTask ? ((AccordTask<?>) t).commandStore.id() : -1;
snapshot.add(new TaskInfo(ifCurrent, commmandStoreId, t));
}
}
}
}

5
src/java/org/apache/cassandra/service/accord/AccordExecutorAbstractLockLoop.java
View File

@ -25,7 +25,6 @@ import accord.api.Agent;
import accord.utils.QuadFunction;
import accord.utils.QuintConsumer;
import org.apache.cassandra.concurrent.DebuggableTask.DebuggableTaskRunner;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import org.apache.cassandra.service.accord.AccordExecutorLoops.LoopTask;
import org.apache.cassandra.utils.concurrent.ConcurrentLinkedStack;
@ -37,9 +36,9 @@ abstract class AccordExecutorAbstractLockLoop extends AccordExecutor
boolean isHeldByExecutor;
boolean shutdown;
AccordExecutorAbstractLockLoop(Lock lock, int executorId, AccordCacheMetrics metrics, Agent agent)
AccordExecutorAbstractLockLoop(Lock lock, int executorId, Agent agent)
{
super(lock, executorId, metrics, agent);
super(lock, executorId, agent);
}
abstract void notifyWork();

5
src/java/org/apache/cassandra/service/accord/AccordExecutorAbstractSemiSyncSubmit.java
View File

@ -23,13 +23,12 @@ import java.util.concurrent.locks.Lock;
import accord.api.Agent;
import accord.utils.QuadFunction;
import accord.utils.QuintConsumer;
import org.apache.cassandra.metrics.AccordCacheMetrics;
abstract class AccordExecutorAbstractSemiSyncSubmit extends AccordExecutorAbstractLockLoop
{
AccordExecutorAbstractSemiSyncSubmit(Lock lock, int executorId, AccordCacheMetrics metrics, Agent agent)
AccordExecutorAbstractSemiSyncSubmit(Lock lock, int executorId, Agent agent)
{
super(lock, executorId, metrics, agent);
super(lock, executorId, agent);
}
abstract void awaitExclusive() throws InterruptedException;

9
src/java/org/apache/cassandra/service/accord/AccordExecutorAsyncSubmit.java
View File

@ -22,7 +22,6 @@ import java.util.concurrent.TimeUnit;
import java.util.function.IntFunction;
import accord.api.Agent;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import org.apache.cassandra.utils.concurrent.LockWithAsyncSignal;
// WARNING: experimental - needs more testing
@ -31,14 +30,14 @@ class AccordExecutorAsyncSubmit extends AccordExecutorAbstractSemiSyncSubmit
private final AccordExecutorLoops loops;
private final LockWithAsyncSignal lock;
public AccordExecutorAsyncSubmit(int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
public AccordExecutorAsyncSubmit(int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
this(new LockWithAsyncSignal(), executorId, mode, threads, name, metrics, agent);
this(new LockWithAsyncSignal(), executorId, mode, threads, name, agent);
}
private AccordExecutorAsyncSubmit(LockWithAsyncSignal lock, int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
private AccordExecutorAsyncSubmit(LockWithAsyncSignal lock, int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
super(lock, executorId, metrics, agent);
super(lock, executorId, agent);
this.lock = lock;
this.loops = new AccordExecutorLoops(mode, threads, name, this::task);
}

9
src/java/org/apache/cassandra/service/accord/AccordExecutorSemiSyncSubmit.java
View File

@ -24,7 +24,6 @@ import java.util.concurrent.locks.ReentrantLock;
import java.util.function.IntFunction;
import accord.api.Agent;
import org.apache.cassandra.metrics.AccordCacheMetrics;
// WARNING: experimental - needs more testing
class AccordExecutorSemiSyncSubmit extends AccordExecutorAbstractSemiSyncSubmit
@ -33,14 +32,14 @@ class AccordExecutorSemiSyncSubmit extends AccordExecutorAbstractSemiSyncSubmit
private final ReentrantLock lock;
private final Condition hasWork;
public AccordExecutorSemiSyncSubmit(int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
public AccordExecutorSemiSyncSubmit(int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
this(new ReentrantLock(), executorId, mode, threads, name, metrics, agent);
this(new ReentrantLock(), executorId, mode, threads, name, agent);
}
private AccordExecutorSemiSyncSubmit(ReentrantLock lock, int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
private AccordExecutorSemiSyncSubmit(ReentrantLock lock, int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
super(lock, executorId, metrics, agent);
super(lock, executorId, agent);
this.lock = lock;
this.hasWork = lock.newCondition();
this.loops = new AccordExecutorLoops(mode, threads, name, this::task);

17
src/java/org/apache/cassandra/service/accord/AccordExecutorSimple.java
View File

@ -27,7 +27,6 @@ import accord.utils.Invariants;
import accord.utils.QuadFunction;
import accord.utils.QuintConsumer;
import org.apache.cassandra.concurrent.ExecutorPlus;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import static org.apache.cassandra.concurrent.ExecutorFactory.Global.executorFactory;
@ -37,26 +36,26 @@ class AccordExecutorSimple extends AccordExecutor
final ReentrantLock lock;
private Task active;
public AccordExecutorSimple(int executorId, String name, AccordCacheMetrics metrics, Agent agent)
public AccordExecutorSimple(int executorId, String name, Agent agent)
{
this(new ReentrantLock(), executorId, name, metrics, agent);
this(new ReentrantLock(), executorId, name, agent);
}
public AccordExecutorSimple(int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
public AccordExecutorSimple(int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
this(new ReentrantLock(), executorId, mode, threads, name, metrics, agent);
this(new ReentrantLock(), executorId, mode, threads, name, agent);
}
private AccordExecutorSimple(ReentrantLock lock, int executorId, String name, AccordCacheMetrics metrics, Agent agent)
private AccordExecutorSimple(ReentrantLock lock, int executorId, String name, Agent agent)
{
super(lock, executorId, metrics, agent);
super(lock, executorId, agent);
this.lock = lock;
this.executor = executorFactory().sequential(name);
}
public AccordExecutorSimple(ReentrantLock lock, int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
public AccordExecutorSimple(ReentrantLock lock, int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
super(lock, executorId, metrics, agent);
super(lock, executorId, agent);
Invariants.requireArgument(threads == 1);
this.lock = lock;
this.executor = executorFactory().sequential(name.apply(0));

13
src/java/org/apache/cassandra/service/accord/AccordExecutorSyncSubmit.java
View File

@ -26,7 +26,6 @@ import java.util.function.IntFunction;
import accord.api.Agent;
import accord.utils.QuadFunction;
import accord.utils.QuintConsumer;
import org.apache.cassandra.metrics.AccordCacheMetrics;
class AccordExecutorSyncSubmit extends AccordExecutorAbstractLockLoop
{
@ -34,19 +33,19 @@ class AccordExecutorSyncSubmit extends AccordExecutorAbstractLockLoop
private final ReentrantLock lock;
private final Condition hasWork;
public AccordExecutorSyncSubmit(int executorId, Mode mode, String name, AccordCacheMetrics metrics, Agent agent)
public AccordExecutorSyncSubmit(int executorId, Mode mode, String name, Agent agent)
{
this(executorId, mode, 1, constant(name), metrics, agent);
this(executorId, mode, 1, constant(name), agent);
}
public AccordExecutorSyncSubmit(int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
public AccordExecutorSyncSubmit(int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
this(new ReentrantLock(), executorId, mode, threads, name, metrics, agent);
this(new ReentrantLock(), executorId, mode, threads, name, agent);
}
private AccordExecutorSyncSubmit(ReentrantLock lock, int executorId, Mode mode, int threads, IntFunction<String> name, AccordCacheMetrics metrics, Agent agent)
private AccordExecutorSyncSubmit(ReentrantLock lock, int executorId, Mode mode, int threads, IntFunction<String> name, Agent agent)
{
super(lock, executorId, metrics, agent);
super(lock, executorId, agent);
this.lock = lock;
this.hasWork = lock.newCondition();
this.loops = new AccordExecutorLoops(mode, threads, name, this::task);

2
src/java/org/apache/cassandra/service/accord/AccordJournal.java
View File

@ -1049,7 +1049,7 @@ public class AccordJournal implements accord.api.Journal, RangeSearcher.Supplier
{
// Since we are iterating in reverse order, we skip the fields that were
// set by entries written later (i.e. already read ones).
if (isChanged(field, flags) && field != CLEANUP)
if (isChanged(field, flags | mask) && field != CLEANUP)
skip(txnId, field, in, userVersion);
else
deserialize(field, in, userVersion);

6
src/java/org/apache/cassandra/service/accord/AccordMessageSink.java
View File

@ -38,6 +38,7 @@ import accord.messages.Reply;
import accord.messages.ReplyContext;
import accord.messages.Request;
import accord.primitives.TxnId;
import accord.utils.async.Cancellable;
import org.apache.cassandra.exceptions.RequestFailureReason;
import org.apache.cassandra.locator.InetAddressAndPort;
import org.apache.cassandra.net.Message;
@ -180,7 +181,7 @@ public class AccordMessageSink implements MessageSink
// TODO (expected): permit bulk send to save esp. on callback registration (and combine records)
@Override
public void send(Node.Id to, Request request, int attempt, AsyncExecutor executor, Callback callback)
public Cancellable send(Node.Id to, Request request, int attempt, AsyncExecutor executor, Callback callback)
{
Verb verb = VerbMapping.getVerb(request);
Preconditions.checkNotNull(verb, "Verb is null for type %s", request.type());
@ -214,8 +215,9 @@ public class AccordMessageSink implements MessageSink
Message<Request> message = Message.out(verb, request, expiresAtNanos);
InetAddressAndPort endpoint = endpointMapper.mappedEndpoint(to);
logger.trace("Sending {} {} to {}", verb, message.payload, endpoint);
callbacks.registerAt(message.id(), executor, callback, to, nowNanos, slowAtNanos, expiresAtNanos, NANOSECONDS);
Cancellable cancellable = callbacks.registerAt(message.id(), executor, callback, to, nowNanos, slowAtNanos, expiresAtNanos, NANOSECONDS);
messaging.send(message, endpoint);
return cancellable;
}
@Override

3
src/java/org/apache/cassandra/service/accord/AccordResponseVerbHandler.java
View File

@ -22,7 +22,6 @@ import java.util.concurrent.TimeUnit;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import accord.coordinate.Timeout;
import accord.impl.RequestCallbacks;
import accord.local.Node;
import accord.messages.Reply;
@ -84,7 +83,7 @@ class AccordResponseVerbHandler<T extends Reply> implements IVerbHandler<T>
private static Throwable convertFailureMessage(RequestFailure failure)
{
return failure.reason == RequestFailureReason.TIMEOUT ?
new Timeout(null, null) :
null :
new RuntimeException(failure.failure);
}

158
src/java/org/apache/cassandra/service/accord/AccordResult.java
View File

@ -21,6 +21,7 @@ package org.apache.cassandra.service.accord;
import java.util.HashSet;
import java.util.Iterator;
import java.util.Set;
import java.util.concurrent.TimeoutException;
import java.util.function.BiConsumer;
import javax.annotation.Nullable;
@ -41,6 +42,7 @@ import org.apache.cassandra.exceptions.ReadFailureException;
import org.apache.cassandra.exceptions.ReadTimeoutException;
import org.apache.cassandra.exceptions.RequestExecutionException;
import org.apache.cassandra.exceptions.RequestFailureException;
import org.apache.cassandra.exceptions.RequestTimeoutException;
import org.apache.cassandra.schema.Schema;
import org.apache.cassandra.schema.TableId;
import org.apache.cassandra.service.accord.api.AccordAgent;
@ -78,11 +80,11 @@ public class AccordResult<V> extends AsyncFuture<V> implements BiConsumer<V, Thr
try
{
if (!awaitUntil(deadlineAtNanos))
accept(null, new Timeout(txnId, null));
tryFailure(new TimeoutException());
}
catch (InterruptedException e)
{
accept(null, e);
tryFailure(e);
}
Throwable fail = fail();
@ -114,88 +116,108 @@ public class AccordResult<V> extends AsyncFuture<V> implements BiConsumer<V, Thr
return timeout;
}
@Override
public void accept(V success, Throwable fail)
public boolean tryFailure(Throwable fail)
{
if (fail != null)
{
RequestExecutionException report;
CoordinationFailed coordinationFailed = findCoordinationFailed(fail);
TxnId txnId = this.txnId;
if (coordinationFailed != null)
{
if (txnId == null && coordinationFailed.txnId() != null)
txnId = coordinationFailed.txnId();
// TODO (expected): increment counters after super.tryFailure
if (isDone())
return false;
if (coordinationFailed instanceof Timeout)
RequestExecutionException report;
CoordinationFailed coordinationFailed = findCoordinationFailed(fail);
TxnId txnId = this.txnId;
if (coordinationFailed != null)
{
if (txnId == null && coordinationFailed.txnId() != null)
txnId = coordinationFailed.txnId();
if (coordinationFailed instanceof Timeout)
{
// Preserve the interop execution created exception if there is one
Throwable maybeWrappedInRequestFailureException = maybeWrappedInRequestFailureException((Timeout)coordinationFailed);
if (maybeWrappedInRequestFailureException instanceof RequestFailureException)
{
// Preserve the interop execution created exception if there is one
Throwable maybeWrappedInRequestFailureException = maybeWrappedInRequestFailureException((Timeout)coordinationFailed);
if (maybeWrappedInRequestFailureException instanceof RequestFailureException)
{
bookkeeping.newTimeout(txnId, keysOrRanges);
report = (RequestFailureException)maybeWrappedInRequestFailureException;
}
else
{
report = bookkeeping.newTimeout(txnId, keysOrRanges);
}
}
else if (coordinationFailed instanceof Preempted)
{
report = bookkeeping.newPreempted(txnId, keysOrRanges);
}
else if (coordinationFailed instanceof Exhausted)
{
report = bookkeeping.newExhausted(txnId, keysOrRanges);
}
else if (isTxnRequest && coordinationFailed instanceof TopologyMismatch)
{
// Excluding bugs topology mismatch can occur because a table was dropped in between creating the txn
// and executing it.
// It could also race with the table stopping/starting being managed by Accord.
// The caller can retry if the table indeed exists and is managed by Accord.
Set<TableId> txnDroppedTables = txnDroppedTables(keysOrRanges);
Tracing.trace("Accord returned topology mismatch: " + coordinationFailed.getMessage());
logger.debug("Accord returned topology mismatch", coordinationFailed);
bookkeeping.markTopologyMismatch();
// Throw IRE in case the caller fails to check if the table still exists
if (!txnDroppedTables.isEmpty())
{
Tracing.trace("Accord txn uses dropped tables {}", txnDroppedTables);
logger.debug("Accord txn uses dropped tables {}", txnDroppedTables);
throw new InvalidRequestException("Accord transaction uses dropped tables");
}
trySuccess((V) RetryWithNewProtocolResult.instance);
return;
bookkeeping.newTimeout(txnId, keysOrRanges);
report = (RequestFailureException)maybeWrappedInRequestFailureException;
}
else
{
report = bookkeeping.newFailed(txnId, keysOrRanges);
report = bookkeeping.newTimeout(txnId, keysOrRanges);
}
// this case happens when a non-timeout exception is seen, and we are unable to move forward
if (txnId != null && txnId.isSyncPoint())
AccordAgent.onFailedBarrier(txnId, fail);
}
else if (coordinationFailed instanceof Preempted)
{
report = bookkeeping.newPreempted(txnId, keysOrRanges);
}
else if (coordinationFailed instanceof Exhausted)
{
report = bookkeeping.newExhausted(txnId, keysOrRanges);
}
else if (isTxnRequest && coordinationFailed instanceof TopologyMismatch)
{
// Excluding bugs topology mismatch can occur because a table was dropped in between creating the txn
// and executing it.
// It could also race with the table stopping/starting being managed by Accord.
// The caller can retry if the table indeed exists and is managed by Accord.
Set<TableId> txnDroppedTables = txnDroppedTables(keysOrRanges);
Tracing.trace("Accord returned topology mismatch: " + coordinationFailed.getMessage());
logger.debug("Accord returned topology mismatch", coordinationFailed);
bookkeeping.markTopologyMismatch();
// Throw IRE in case the caller fails to check if the table still exists
if (!txnDroppedTables.isEmpty())
{
Tracing.trace("Accord txn uses dropped tables {}", txnDroppedTables);
logger.debug("Accord txn uses dropped tables {}", txnDroppedTables);
throw new InvalidRequestException("Accord transaction uses dropped tables");
}
trySuccess((V) RetryWithNewProtocolResult.instance);
return false;
}
else
{
logger.error("Unexpected exception", fail);
JVMStabilityInspector.inspectThrowable(fail);
report = bookkeeping.newFailed(txnId, keysOrRanges);
}
report.addSuppressed(fail);
tryFailure(report);
// this case happens when a non-timeout exception is seen, and we are unable to move forward
if (txnId != null && txnId.isSyncPoint())
AccordAgent.onFailedBarrier(txnId, fail);
}
else if (fail instanceof RequestTimeoutException || fail instanceof TimeoutException)
{
report = bookkeeping.newTimeout(txnId, keysOrRanges);
}
else
{
if (success == RetryWithNewProtocolResult.instance)
{
bookkeeping.markRetryDifferentSystem();
Tracing.trace("Got retry different system error from Accord, will retry");
}
trySuccess(success);
logger.error("Unexpected exception", fail);
JVMStabilityInspector.inspectThrowable(fail);
report = bookkeeping.newFailed(txnId, keysOrRanges);
}
report.addSuppressed(fail);
if (!super.tryFailure(report))
return false;
bookkeeping.markElapsedNanos(nanoTime() - startedAtNanos);
return true;
}
@Override
protected boolean trySuccess(V success)
{
if (!super.trySuccess(success))
return false;
bookkeeping.markElapsedNanos(nanoTime() - startedAtNanos);
if (success == RetryWithNewProtocolResult.instance)
{
bookkeeping.markRetryDifferentSystem();
Tracing.trace("Got retry different system error from Accord, will retry");
}
return true;
}
@Override
public void accept(V success, Throwable fail)
{
if (fail == null) trySuccess(success);
else tryFailure(fail);
}
@Override
@ -204,7 +226,7 @@ public class AccordResult<V> extends AsyncFuture<V> implements BiConsumer<V, Thr
if (super.awaitUntil(nanoTimeDeadline))
return true;
accept(null, new Timeout(null, null));
tryFailure(new TimeoutException());
return false;
}

13
src/java/org/apache/cassandra/service/accord/AccordService.java
View File

@ -39,6 +39,7 @@ import javax.annotation.concurrent.GuardedBy;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.primitives.Ints;
import org.apache.cassandra.metrics.AccordReplicaMetrics;
import org.apache.cassandra.utils.Clock;
import org.apache.cassandra.utils.concurrent.ImmediateFuture;
import org.slf4j.Logger;
@ -243,7 +244,9 @@ public class AccordService implements IAccordService, Shutdownable
private static final IAccordService NOOP_SERVICE = new NoOpAccordService();
private static volatile IAccordService instance = null;
// TODO (expected): wrap this in an inner class that is statically initialised and final
// tests can specify a DelegatingService if they want to override
private static IAccordService instance;
@VisibleForTesting
public static void unsafeSetNewAccordService(IAccordService service)
@ -297,6 +300,8 @@ public class AccordService implements IAccordService, Shutdownable
as.startup();
instance = as;
AccordReplicaMetrics.touch();
replayJournal(as);
as.finishInitialization();
@ -701,6 +706,12 @@ public class AccordService implements IAccordService, Shutdownable
return ((AccordCommandStores)node.commandStores()).executors();
}
@Override
public boolean isEnabled()
{
return true;
}
@Override
public @Nonnull IAccordResult<TxnResult> coordinateAsync(long minEpoch, @Nonnull Txn txn, @Nonnull ConsistencyLevel consistencyLevel, @Nonnull Dispatcher.RequestTime requestTime, long minHlc)
{

15
src/java/org/apache/cassandra/service/accord/IAccordService.java
View File

@ -106,6 +106,7 @@ public interface IAccordService
IAccordResult<V> addCallback(BiConsumer<? super V, Throwable> callback);
}
boolean isEnabled();
long currentEpoch();
void setCacheSize(long kb);
@ -248,6 +249,12 @@ public interface IAccordService
throw new UnsupportedOperationException("No accord transaction should be executed when accord.enabled = false in cassandra.yaml");
}
@Override
public boolean isEnabled()
{
return false;
}
@Override
public long currentEpoch()
{
@ -383,7 +390,7 @@ public interface IAccordService
class DelegatingAccordService implements IAccordService
{
protected final IAccordService delegate;
protected IAccordService delegate;
public DelegatingAccordService(IAccordService delegate)
{
@ -439,6 +446,12 @@ public interface IAccordService
return delegate.executors();
}
@Override
public boolean isEnabled()
{
return delegate.isEnabled();
}
@Nonnull
@Override
public IAccordResult<TxnResult> coordinateAsync(long minEpoch, @Nonnull Txn txn, @Nonnull ConsistencyLevel consistencyLevel, RequestTime requestTime, long minHlc)

81
src/java/org/apache/cassandra/service/accord/api/AccordAgent.java
View File

@ -30,7 +30,7 @@ import org.slf4j.LoggerFactory;
import accord.api.Agent;
import accord.api.CoordinatorEventListener;
import accord.api.LocalEventListener;
import accord.api.ReplicaEventListener;
import accord.api.ProgressLog.BlockedUntil;
import accord.api.RoutingKey;
import accord.api.Tracing;
@ -62,7 +62,8 @@ import accord.utils.async.AsyncResult;
import accord.utils.async.AsyncResults;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.exceptions.RequestTimeoutException;
import org.apache.cassandra.metrics.AccordMetrics;
import org.apache.cassandra.metrics.AccordCoordinatorMetrics;
import org.apache.cassandra.metrics.AccordReplicaMetrics;
import org.apache.cassandra.net.ResponseContext;
import org.apache.cassandra.service.accord.AccordService;
import org.apache.cassandra.service.accord.AccordTracing;
@ -71,11 +72,13 @@ import org.apache.cassandra.service.accord.txn.TxnQuery;
import org.apache.cassandra.service.accord.txn.TxnRead;
import org.apache.cassandra.utils.Clock;
import org.apache.cassandra.utils.JVMStabilityInspector;
import org.apache.cassandra.utils.NoSpamLogger;
import static accord.primitives.Routable.Domain.Key;
import static accord.utils.SortedArrays.SortedArrayList.ofSorted;
import static java.util.concurrent.TimeUnit.MICROSECONDS;
import static java.util.concurrent.TimeUnit.MILLISECONDS;
import static java.util.concurrent.TimeUnit.MINUTES;
import static java.util.concurrent.TimeUnit.NANOSECONDS;
import static java.util.concurrent.TimeUnit.SECONDS;
import static org.apache.cassandra.config.DatabaseDescriptor.getAccordScheduleDurabilityTxnIdLag;
@ -94,6 +97,7 @@ import static org.apache.cassandra.utils.Clock.Global.nanoTime;
public class AccordAgent implements Agent
{
private static final Logger logger = LoggerFactory.getLogger(AccordAgent.class);
private static final NoSpamLogger noSpamLogger = NoSpamLogger.getLogger(logger, 1L, MINUTES);
private static BiConsumer<TxnId, Throwable> onFailedBarrier;
public static void setOnFailedBarrier(BiConsumer<TxnId, Throwable> newOnFailedBarrier) { onFailedBarrier = newOnFailedBarrier; }
@ -224,17 +228,20 @@ public class AccordAgent implements Agent
@Override
public CoordinatorEventListener coordinatorEvents()
{
return AccordMetrics.Listener.instance;
return AccordCoordinatorMetrics.Listener.instance;
}
@Override
public LocalEventListener localEvents()
public ReplicaEventListener replicaEvents()
{
return AccordMetrics.Listener.instance;
return AccordReplicaMetrics.Listener.instance;
}
private static final long ONE_SECOND = SECONDS.toMicros(1L);
private static final long ONE_MINUTE = MINUTES.toMicros(1L);
@Override
public long slowCoordinatorDelay(Node node, SafeCommandStore safeStore, TxnId txnId, TimeUnit units, int retryCount)
public long slowCoordinatorDelay(Node node, SafeCommandStore safeStore, TxnId txnId, TimeUnit units, int attempt)
{
SafeCommand safeCommand = safeStore.unsafeGetNoCleanup(txnId);
Invariants.nonNull(safeCommand);
@ -242,29 +249,48 @@ public class AccordAgent implements Agent
Command command = safeCommand.current();
Invariants.nonNull(command);
// TODO (expected): make this a configurable calculation on normal request latencies (like ContentionStrategy)
long nowMicros = MILLISECONDS.toMicros(Clock.Global.currentTimeMillis());
long mostRecentStart = mostRecentStart(command, nowMicros);
long waitMicros = recover(txnId).computeWait(attempt, MICROSECONDS);
long startTime = mostRecentStart + waitMicros;
if (startTime < nowMicros)
{
// TODO (expected): support no waiting here
if (attempt == 1)
return 1;
startTime = nowMicros + waitMicros/2;
}
RoutingKey homeKey = command.route().homeKey();
Shard shard = node.topology().forEpochIfKnown(homeKey, command.txnId().epoch());
// TODO (expected): make this a configurable calculation on normal request latencies (like ContentionStrategy)
long nowMicros = MILLISECONDS.toMicros(Clock.Global.currentTimeMillis());
long oneSecond = SECONDS.toMicros(1L);
long promisedHlc = command.promised().hlc();
if (promisedHlc > nowMicros + TimeUnit.MINUTES.toMicros(1))
promisedHlc = 0;
long mostRecentStart = Math.max(command.txnId().hlc(), promisedHlc);
long waitMicros = recover(txnId).computeWait(retryCount, MICROSECONDS);
if (mostRecentStart > nowMicros + SECONDS.toMicros(1L))
logger.warn("max({},{})>{}", command.txnId(), command.promised(), nowMicros);
long startTime = mostRecentStart + waitMicros;
if (startTime < nowMicros)
startTime = nowMicros + waitMicros/2;
startTime = nonClashingStartTime(startTime, shard == null ? null : shard.nodes, node.id(), oneSecond, random);
startTime = nonClashingStartTime(startTime, shard == null ? null : shard.nodes, node.id(), ONE_SECOND, random);
long delayMicros = Math.max(1, startTime - nowMicros);
Invariants.require(delayMicros < TimeUnit.HOURS.toMicros(1L), "unexpectedly long coordination recovery delay proposed: %d (start %d, now %d)", delayMicros, startTime, nowMicros, command.txnId(), command.promised());
return units.convert(delayMicros, MICROSECONDS);
}
private static long mostRecentStart(Command command, long nowMicros)
{
// TODO (expected): make this a configurable calculation on normal request latencies (like ContentionStrategy)
long promisedHlc = command.promised().hlc();
if (promisedHlc > nowMicros + ONE_MINUTE)
promisedHlc = 0;
long result = Math.max(command.txnId().hlc(), promisedHlc);
if (result > nowMicros + ONE_SECOND)
noSpamLogger.warn("max({},{})>{}", command.txnId(), command.promised(), nowMicros);
return result;
}
@Override
public boolean isSlowCoordinator(long elapsed, TimeUnit units, TxnId txnId, int attempt)
{
long maxWait = recover(txnId).computeMaxWait(attempt, units);
return elapsed >= maxWait;
}
@VisibleForTesting
public static long nonClashingStartTime(long startTime, SortedList<Node.Id> nodes, Node.Id id, long granularity, RandomSource random)
{
@ -291,7 +317,18 @@ public class AccordAgent implements Agent
@Override
public long slowReplicaDelay(Node node, SafeCommandStore safeStore, TxnId txnId, int attempt, BlockedUntil blockedUntil, TimeUnit units)
{
return fetch(txnId).computeWait(attempt, units);
Command command = Invariants.nonNull(safeStore.unsafeGetNoCleanup(txnId).current());
long nowMicros = MILLISECONDS.toMicros(Clock.Global.currentTimeMillis());
long mostRecentStart = mostRecentStart(command, nowMicros);
long waitMicros = fetch(txnId).computeWait(attempt, units);
long startTime = mostRecentStart + waitMicros;
if (startTime < nowMicros)
{
// TODO (expected): support no waiting here
if (attempt == 1) return 1;
else return waitMicros/2;
}
return waitMicros;
}
@Override

10
src/java/org/apache/cassandra/service/accord/events/CacheEvents.java
View File

@ -40,7 +40,7 @@ public abstract class CacheEvents extends Event
// instance
@DataAmount(DataAmount.BYTES)
public long instanceAllocated;
public long instanceStatsQueries, instanceStatsHits, instanceStatsMisses;
public long instanceStatsHits, instanceStatsMisses;
@Percentage
public double instanceStatsHitRate;
@ -50,17 +50,11 @@ public abstract class CacheEvents extends Event
public long globalCapacity, globalAllocated;
public int globalSize, globalReferenced, globalUnreferenced;
public long globalStatsQueries, globalStatsHits, globalStatsMisses;
@Percentage
public double globalStatsHitRate;
@Percentage
public double globalFree;
public void update()
{
instanceStatsHitRate = 1D - (instanceStatsHits / (double) instanceStatsQueries);
globalStatsHitRate = 1D - (globalStatsHits / (double) globalStatsQueries);
instanceStatsHitRate = 1D - (instanceStatsHits / (double) (instanceStatsHits + instanceStatsMisses));
globalFree = 1.0D - (globalAllocated / (double) globalCapacity);
}

4
src/java/org/apache/cassandra/service/accord/interop/AccordInteropAdapter.java
View File

@ -119,8 +119,8 @@ public class AccordInteropAdapter extends TxnAdapter
return false;
Topologies all = execution(node, any, sendTo, selectSendTo, fullRoute, txnId, executeAt);
new AccordInteropPersist(node, executor, all, txnId, require, ballot, txn, executeAt, deps, writes, result, fullRoute, consistencyLevel, CoordinationFlags.none(), informDurableOnDone, callback)
.start(Minimal, any, writes, result);
new AccordInteropPersist(node, executor, all, txnId, require, ballot, txn, executeAt, deps, writes, result, fullRoute, consistencyLevel, CoordinationFlags.none(), informDurableOnDone, Minimal, callback)
.start();
return true;
}
}

8
src/java/org/apache/cassandra/service/accord/interop/AccordInteropApply.java
View File

@ -166,14 +166,6 @@ public class AccordInteropApply extends Apply implements LocalListeners.ComplexL
node.reply(replyTo, replyContext, ApplyReply.Applied, null);
}
@Override
public ApplyReply reduce(ApplyReply r1, ApplyReply r2)
{
return r1 == null || r2 == null
? r1 == null ? r2 : r1
: r1.compareTo(r2) >= 0 ? r1 : r2;
}
@Override
protected void acceptInternal(ApplyReply reply, Throwable failure)
{

22
src/java/org/apache/cassandra/service/accord/interop/AccordInteropExecution.java
View File

@ -31,7 +31,6 @@ import accord.api.Data;
import accord.api.Result;
import accord.coordinate.CoordinationAdapter;
import accord.coordinate.ExecuteFlag.CoordinationFlags;
import accord.coordinate.Timeout;
import accord.local.Node;
import accord.local.Node.Id;
import accord.local.SequentialAsyncExecutor;
@ -67,8 +66,6 @@ import org.apache.cassandra.db.partitions.PartitionIterator;
import org.apache.cassandra.db.rows.RowIterator;
import org.apache.cassandra.dht.Range;
import org.apache.cassandra.dht.Token;
import org.apache.cassandra.exceptions.ReadFailureException;
import org.apache.cassandra.exceptions.ReadTimeoutException;
import org.apache.cassandra.locator.EndpointsForToken;
import org.apache.cassandra.locator.InetAddressAndPort;
import org.apache.cassandra.locator.Replica;
@ -391,28 +388,11 @@ public class AccordInteropExecution implements ReadCoordinator
}
else
{
callback.accept(null, maybeWrapRequestFailureException(failure));
callback.accept(null, failure);
}
});
}
/**
* Interop should expose these exceptions as the appropriate Accord types so AccordService
* knows how to handle them
*/
private Throwable maybeWrapRequestFailureException(Throwable failure)
{
Throwable toCheck = failure;
do
{
// TODO (required): There are probably more exceptions that will have this issue of wanting
// to be turned into the top level exception sent back to the client
if (toCheck instanceof ReadTimeoutException || toCheck instanceof ReadFailureException)
return new Timeout(txnId, route.homeKey(), failure);
} while ((toCheck = toCheck.getCause()) != null);
return failure;
}
private AsyncChain<Data> executeUnrecoverableRepairUpdate()
{
return AsyncChains.ofCallable(Stage.ACCORD_MIGRATION.executor(), () -> {

22
src/java/org/apache/cassandra/service/accord/interop/AccordInteropPersist.java
View File

@ -40,6 +40,7 @@ import accord.primitives.TxnId;
import accord.primitives.Writes;
import accord.topology.Topologies;
import accord.utils.Invariants;
import accord.utils.UnhandledEnum;
import org.apache.cassandra.db.ConsistencyLevel;
import org.apache.cassandra.utils.Throwables;
@ -111,9 +112,9 @@ public class AccordInteropPersist extends Persist
private final ConsistencyLevel consistencyLevel;
private CallbackHolder callback;
public AccordInteropPersist(Node node, SequentialAsyncExecutor executor, Topologies topologies, TxnId txnId, Route<?> sendTo, Ballot ballot, Txn txn, Timestamp executeAt, Deps deps, Writes writes, Result result, FullRoute<?> fullRoute, ConsistencyLevel consistencyLevel, ExecuteFlag.CoordinationFlags flags, boolean informDurableOnDone, BiConsumer<? super Result, Throwable> clientCallback)
public AccordInteropPersist(Node node, SequentialAsyncExecutor executor, Topologies topologies, TxnId txnId, Route<?> sendTo, Ballot ballot, Txn txn, Timestamp executeAt, Deps deps, Writes writes, Result result, FullRoute<?> fullRoute, ConsistencyLevel consistencyLevel, ExecuteFlag.CoordinationFlags flags, boolean informDurableOnDone, Apply.Kind applyKind, BiConsumer<? super Result, Throwable> clientCallback)
{
super(node, executor, topologies, txnId, ballot, sendTo, txn, executeAt, deps, writes, result, fullRoute, flags, informDurableOnDone, AccordInteropApply.FACTORY);
super(node, executor, topologies, txnId, ballot, sendTo, txn, executeAt, deps, writes, result, fullRoute, flags, informDurableOnDone, AccordInteropApply.FACTORY, applyKind);
Invariants.requireArgument(consistencyLevel == ConsistencyLevel.QUORUM || consistencyLevel == ConsistencyLevel.ALL || consistencyLevel == ConsistencyLevel.SERIAL || consistencyLevel == ConsistencyLevel.ONE);
this.consistencyLevel = consistencyLevel;
registerClientCallback(result, clientCallback);
@ -127,10 +128,10 @@ public class AccordInteropPersist extends Persist
case ONE: // Can safely upgrade ONE to QUORUM/SERIAL to get a synchronous commit
case SERIAL:
case QUORUM:
callback = new CallbackHolder(new QuorumTracker(topologies), result, clientCallback);
callback = new CallbackHolder(new QuorumTracker(tracker.topologies()), result, clientCallback);
break;
case ALL:
callback = new CallbackHolder(new AllTracker(topologies), result, clientCallback);
callback = new CallbackHolder(new AllTracker(tracker.topologies()), result, clientCallback);
break;
default:
throw new IllegalArgumentException("Unhandled consistency level: " + consistencyLevel);
@ -141,8 +142,9 @@ public class AccordInteropPersist extends Persist
public void onSuccess(Node.Id from, Apply.ApplyReply reply)
{
super.onSuccess(from, reply);
switch (reply)
switch (reply.kind)
{
case InsufficientEpochs: throw UnhandledEnum.invalid(reply.kind);
case Redundant:
case Applied:
callback.recordSuccess(from);
@ -151,19 +153,27 @@ public class AccordInteropPersist extends Persist
// On insufficient Persist will send a commit with the missing information
// which will allow a final response to be returned later that could be successful
return;
default: throw new IllegalArgumentException("Unhandled apply response " + reply);
default: throw UnhandledEnum.unknown(reply.kind);
}
}
@Override
public void start()
{
super.start();
}
@Override
public void onFailure(Node.Id from, Throwable failure)
{
callback.recordFailure(from, failure);
super.onFailure(from, failure);
}
@Override
public boolean onCallbackFailure(Node.Id from, Throwable failure)
{
super.onCallbackFailure(from, failure);
return callback.recordCallbackFailure(failure);
}
}

6
src/java/org/apache/cassandra/service/accord/repair/RequiredResponseTracker.java
View File

@ -58,6 +58,12 @@ public class RequiredResponseTracker extends SimpleTracker<RequiredResponseTrack
{
return !outstandingResponses.isEmpty() ? Fail : NoChange;
}
@Override
public String summarise()
{
return (shard.rf - outstandingResponses.size()) + "/" + shard.rf;
}
}
public RequiredResponseTracker(Set<Node.Id> requiredResponses, Topologies topologies)

62
src/java/org/apache/cassandra/service/accord/serializers/ApplySerializers.java
View File

@ -23,6 +23,7 @@ import java.io.IOException;
import accord.api.Result;
import accord.coordinate.ExecuteFlag.ExecuteFlags;
import accord.messages.Apply;
import accord.messages.Apply.ApplyReply;
import accord.primitives.Ballot;
import accord.primitives.FullRoute;
import accord.primitives.PartialDeps;
@ -31,47 +32,27 @@ import accord.primitives.Route;
import accord.primitives.Timestamp;
import accord.primitives.TxnId;
import accord.primitives.Writes;
import accord.utils.Invariants;
import accord.utils.VIntCoding;
import org.apache.cassandra.db.TypeSizes;
import org.apache.cassandra.io.UnversionedSerializer;
import org.apache.cassandra.io.util.DataInputPlus;
import org.apache.cassandra.io.util.DataOutputPlus;
import org.apache.cassandra.service.accord.serializers.CommandSerializers.ExecuteAtSerializer;
import static accord.primitives.Txn.Kind.Write;
public class ApplySerializers
{
private static final UnversionedSerializer<Apply.Kind> kind = new UnversionedSerializer<>()
{
public void serialize(Apply.Kind kind, DataOutputPlus out) throws IOException
{
Invariants.requireArgument(kind == Apply.Kind.Maximal || kind == Apply.Kind.Minimal);
out.writeBoolean(kind == Apply.Kind.Maximal);
}
public Apply.Kind deserialize(DataInputPlus in) throws IOException
{
return in.readBoolean() ? Apply.Kind.Maximal : Apply.Kind.Minimal;
}
public long serializedSize(Apply.Kind t)
{
return TypeSizes.BOOL_SIZE;
}
};
public abstract static class ApplySerializer<A extends Apply> extends TxnRequestSerializer<A>
{
private static final EncodeAsVInt32<Apply.Kind> kinds = EncodeAsVInt32.of(Apply.Kind.class);
@Override
public void serializeBody(A apply, DataOutputPlus out, Version version) throws IOException
{
CommandSerializers.ballot.serialize(apply.ballot, out);
out.writeVInt(apply.minEpoch - apply.waitForEpoch);
out.writeUnsignedVInt(apply.maxEpoch - apply.minEpoch);
kind.serialize(apply.kind, out);
kinds.serialize(apply.kind, out);
ExecuteAtSerializer.serialize(apply.txnId, apply.executeAt, out);
DepsSerializers.partialDeps.serialize(apply.deps(), out);
CommandSerializers.nullablePartialTxn.serialize(apply.txn(), out, version);
@ -91,7 +72,7 @@ public class ApplySerializers
long minEpoch = waitForEpoch + in.readVInt();
long maxEpoch = minEpoch + in.readUnsignedVInt();
return deserializeApply(txnId, ballot, scope, minEpoch, waitForEpoch, maxEpoch,
kind.deserialize(in),
kinds.deserialize(in),
ExecuteAtSerializer.deserialize(txnId, in),
DepsSerializers.partialDeps.deserialize(in),
CommandSerializers.nullablePartialTxn.deserialize(in, version),
@ -107,7 +88,7 @@ public class ApplySerializers
return CommandSerializers.ballot.serializedSize(apply.ballot)
+ TypeSizes.sizeofVInt(apply.minEpoch - apply.waitForEpoch)
+ TypeSizes.sizeofUnsignedVInt(apply.maxEpoch - apply.minEpoch)
+ kind.serializedSize(apply.kind)
+ kinds.serializedSize(apply.kind)
+ ExecuteAtSerializer.serializedSize(apply.txnId, apply.executeAt)
+ DepsSerializers.partialDeps.serializedSize(apply.deps())
+ CommandSerializers.nullablePartialTxn.serializedSize(apply.txn(), version)
@ -128,26 +109,37 @@ public class ApplySerializers
}
};
public static final UnversionedSerializer<Apply.ApplyReply> reply = new UnversionedSerializer<>()
public static final IVersionedSerializer<ApplyReply> reply = new ReplySerializer();
public static final class ReplySerializer implements IVersionedSerializer<ApplyReply>
{
private final Apply.ApplyReply[] replies = Apply.ApplyReply.values();
private static final EncodeAsVInt32<ApplyReply.Kind> kinds = EncodeAsVInt32.of(ApplyReply.Kind.class);
@Override
public void serialize(Apply.ApplyReply t, DataOutputPlus out) throws IOException
public void serialize(ApplyReply t, DataOutputPlus out, Version version) throws IOException
{
out.writeByte(t.ordinal());
kinds.serialize(t.kind, out);
if (t.kind == ApplyReply.Kind.InsufficientEpochs)
out.writeUnsignedVInt(t.minEpoch());
}
@Override
public Apply.ApplyReply deserialize(DataInputPlus in) throws IOException
public ApplyReply deserialize(DataInputPlus in, Version version) throws IOException
{
return replies[in.readByte()];
ApplyReply.Kind kind = kinds.deserialize(in);
if (kind != ApplyReply.Kind.InsufficientEpochs)
return ApplyReply.lookupByKind(kind);
long minEpoch = in.readUnsignedVInt();
return new ApplyReply(kind, minEpoch);
}
@Override
public long serializedSize(Apply.ApplyReply t)
public long serializedSize(ApplyReply t, Version version)
{
return 1;
long size = kinds.serializedSize(t.kind);
if (t.kind == ApplyReply.Kind.InsufficientEpochs)
size += VIntCoding.sizeOfUnsignedVInt(t.minEpoch());
return size;
}
};
}
}

17
src/java/org/apache/cassandra/service/accord/serializers/AwaitSerializers.java
View File

@ -20,7 +20,6 @@ package org.apache.cassandra.service.accord.serializers;
import java.io.IOException;
import accord.api.ProgressLog.BlockedUntil;
import accord.messages.Await;
import accord.messages.Await.AsyncAwaitComplete;
import accord.messages.Await.AwaitOk;
@ -42,19 +41,19 @@ public class AwaitSerializers
public static final UnversionedSerializer<Await> request = new RequestSerializer<>()
{
@Override
public Await deserialize(TxnId txnId, Participants<?> scope, BlockedUntil blockedUntil, boolean notifyProgressLog, long minAwaitEpoch, long maxAwaitEpoch, int callbackId, DataInputPlus in)
public Await deserialize(TxnId txnId, Participants<?> scope, Await.Until awaitUntil, boolean notifyProgressLog, long minAwaitEpoch, long maxAwaitEpoch, int callbackId, DataInputPlus in)
{
return Await.SerializerSupport.create(txnId, scope, blockedUntil, notifyProgressLog, minAwaitEpoch, maxAwaitEpoch, callbackId);
return Await.SerializerSupport.create(txnId, scope, awaitUntil, notifyProgressLog, minAwaitEpoch, maxAwaitEpoch, callbackId);
}
};
public static final UnversionedSerializer<RecoverAwait> recoverRequest = new RequestSerializer<>()
{
@Override
public RecoverAwait deserialize(TxnId txnId, Participants<?> scope, BlockedUntil blockedUntil, boolean notifyProgressLog, long minAwaitEpoch, long maxAwaitEpoch, int callbackId, DataInputPlus in) throws IOException
public RecoverAwait deserialize(TxnId txnId, Participants<?> scope, Await.Until awaitUntil, boolean notifyProgressLog, long minAwaitEpoch, long maxAwaitEpoch, int callbackId, DataInputPlus in) throws IOException
{
TxnId recoverId = CommandSerializers.txnId.deserialize(in);
return RecoverAwait.SerializerSupport.create(txnId, scope, blockedUntil, notifyProgressLog, minAwaitEpoch, maxAwaitEpoch, callbackId, recoverId);
return RecoverAwait.SerializerSupport.create(txnId, scope, awaitUntil, notifyProgressLog, minAwaitEpoch, maxAwaitEpoch, callbackId, recoverId);
}
@Override
@ -73,14 +72,14 @@ public class AwaitSerializers
static abstract class RequestSerializer<A extends Await> implements UnversionedSerializer<A>
{
abstract A deserialize(TxnId txnId, Participants<?> scope, BlockedUntil blockedUntil, boolean notifyProgressLog, long minAwaitEpoch, long maxAwaitEpoch, int callbackId, DataInputPlus in) throws IOException;
abstract A deserialize(TxnId txnId, Participants<?> scope, Await.Until awaitUntil, boolean notifyProgressLog, long minAwaitEpoch, long maxAwaitEpoch, int callbackId, DataInputPlus in) throws IOException;
@Override
public void serialize(A await, DataOutputPlus out) throws IOException
{
CommandSerializers.txnId.serialize(await.txnId, out);
KeySerializers.participants.serialize(await.scope, out);
out.writeByte((await.blockedUntil.ordinal() << 1) | (await.notifyProgressLog ? 1 : 0));
out.writeByte((await.until.ordinal() << 1) | (await.notifyProgressLog ? 1 : 0));
out.writeUnsignedVInt(await.maxAwaitEpoch - await.txnId.epoch());
out.writeUnsignedVInt(await.maxAwaitEpoch - await.minAwaitEpoch);
out.writeUnsignedVInt32(await.callbackId + 1);
@ -93,13 +92,13 @@ public class AwaitSerializers
TxnId txnId = CommandSerializers.txnId.deserialize(in);
Participants<?> scope = KeySerializers.participants.deserialize(in);
int blockedAndNotify = in.readByte();
BlockedUntil blockedUntil = BlockedUntil.forOrdinal(blockedAndNotify >>> 1);
Await.Until awaitUntil = Await.Until.forOrdinal(blockedAndNotify >>> 1);
boolean notifyProgressLog = (blockedAndNotify & 1) == 1;
long maxAwaitEpoch = in.readUnsignedVInt() + txnId.epoch();
long minAwaitEpoch = maxAwaitEpoch - in.readUnsignedVInt();
int callbackId = in.readUnsignedVInt32() - 1;
Invariants.require(callbackId >= -1);
return deserialize(txnId, scope, blockedUntil, notifyProgressLog, minAwaitEpoch, maxAwaitEpoch, callbackId, in);
return deserialize(txnId, scope, awaitUntil, notifyProgressLog, minAwaitEpoch, maxAwaitEpoch, callbackId, in);
}
@Override

31
src/java/org/apache/cassandra/service/accord/serializers/DepsSerializers.java
View File

@ -72,12 +72,12 @@ public class DepsSerializers
static final int KEYS_BY_TXNID = 0x1;
static final int RANGES_BY_TXNID = 0x2;
static final int FLAGS_SIZE = VIntCoding.sizeOfUnsignedVInt(KEYS_BY_TXNID | RANGES_BY_TXNID);
final boolean forceByTxnId;
final boolean byTxnId;
protected final UnversionedSerializer<Range> tokenRange;
public AbstractDepsSerializer(boolean forceByTxnId, UnversionedSerializer<Range> tokenRange)
public AbstractDepsSerializer(boolean byTxnId, UnversionedSerializer<Range> tokenRange)
{
this.forceByTxnId = forceByTxnId;
this.byTxnId = byTxnId;
this.tokenRange = tokenRange;
}
@ -86,8 +86,8 @@ public class DepsSerializers
@Override
public void serialize(D deps, DataOutputPlus out) throws IOException
{
boolean keysByTxnId = forceByTxnId || deps.keyDeps.hasByTxnId();
boolean rangesByTxnId = forceByTxnId || deps.rangeDeps.hasByTxnId();
boolean keysByTxnId = byTxnId;
boolean rangesByTxnId = byTxnId;
out.writeUnsignedVInt32((keysByTxnId ? KEYS_BY_TXNID : 0) | (rangesByTxnId ? RANGES_BY_TXNID : 0));
{
KeyDeps keyDeps = deps.keyDeps;
@ -109,7 +109,7 @@ public class DepsSerializers
{
out.writeUnsignedVInt32(xtoy.length);
if ((xCount <= 1 || yCount <= 1) && (xtoy.length == xCount + yCount || xCount == 0 || yCount == 0))
if ((xCount <= 1 || yCount <= 1) && (xtoy.length == (xCount == 1 ? 1 + yCount : 2 * xCount) || xCount == 0 || yCount == 0))
{
// no point serializing as can be directly inferred
if (Invariants.isParanoid())
@ -172,7 +172,7 @@ public class DepsSerializers
int length = in.readUnsignedVInt32();
int[] xtoy = new int[length];
if ((xCount <= 1 || yCount <= 1) && (xtoy.length == xCount + yCount || xCount == 0 || yCount == 0))
if ((xCount <= 1 || yCount <= 1) && (xtoy.length == (xCount == 1 ? 1 + yCount : 2 * xCount) || xCount == 0 || yCount == 0))
{
// no point serializing as can be directly inferred
if (xCount == 1)
@ -207,8 +207,8 @@ public class DepsSerializers
@Override
public long serializedSize(D deps)
{
boolean keysByTxnId = forceByTxnId || deps.keyDeps.hasByTxnId();
boolean rangesByTxnId = forceByTxnId || deps.rangeDeps.hasByTxnId();
boolean keysByTxnId = byTxnId;
boolean rangesByTxnId = byTxnId;
long size = FLAGS_SIZE;
{
KeyDeps keyDeps = deps.keyDeps;
@ -230,7 +230,7 @@ public class DepsSerializers
private static long serializedPackedXtoYSize(int[] xtoy, int xCount, int yCount)
{
long size = VIntCoding.sizeOfUnsignedVInt(xtoy.length);
if ((xCount <= 1 || yCount <= 1) && (xtoy.length == xCount + yCount || xCount == 0 || yCount == 0))
if ((xCount <= 1 || yCount <= 1) && (xtoy.length == (xCount == 1 ? 1 + yCount : 2 * xCount) || xCount == 0 || yCount == 0))
{
// no point serializing as can be directly inferred
}
@ -274,9 +274,9 @@ public class DepsSerializers
static class DepsSerializer extends AbstractDepsSerializer<Deps>
{
public DepsSerializer(boolean preferByTxnId, UnversionedSerializer<Range> tokenRange)
public DepsSerializer(boolean byTxnId, UnversionedSerializer<Range> tokenRange)
{
super(preferByTxnId, tokenRange);
super(byTxnId, tokenRange);
}
@Override
@ -292,7 +292,7 @@ public class DepsSerializers
final UnversionedSerializer<Range> tokenRange;
final DepsSerializer deps;
final UnversionedSerializer<Deps> nullableDeps;
final PartialDepsSerializer partialDeps;
final PartialDepsSerializer partialDeps; // currently happens to be byId but doesn't rely upon it
final PartialDepsSerializer partialDepsById;
final UnversionedSerializer<PartialDeps> nullablePartialDeps;
@ -301,10 +301,9 @@ public class DepsSerializers
this.tokenRange = tokenRange;
this.deps = new DepsSerializer(false, tokenRange);
this.nullableDeps = NullableSerializer.wrap(deps);
this.partialDeps = new PartialDepsSerializer(false, tokenRange);
this.partialDepsById = new PartialDepsSerializer(true, tokenRange);
this.partialDeps = new PartialDepsSerializer(true, tokenRange);
this.partialDepsById = partialDeps;
this.nullablePartialDeps = NullableSerializer.wrap(partialDeps);
}
}
}

23
src/java/org/apache/cassandra/service/accord/serializers/FetchSerializers.java
View File

@ -26,6 +26,7 @@ import accord.impl.AbstractFetchCoordinator.FetchResponse;
import accord.messages.ReadData.CommitOrReadNack;
import accord.messages.ReadData.ReadReply;
import accord.primitives.Ranges;
import accord.utils.VIntCoding;
import org.apache.cassandra.db.TypeSizes;
import org.apache.cassandra.io.UnversionedSerializer;
import org.apache.cassandra.io.util.DataInputPlus;
@ -76,7 +77,6 @@ public class FetchSerializers
public static final UnversionedSerializer<ReadReply> reply = new UnversionedSerializer<>()
{
final CommitOrReadNack[] nacks = CommitOrReadNack.values();
final UnversionedSerializer<Data> streamDataSerializer = CastingSerializer.create(StreamData.class, StreamData.serializer);
@Override
@ -84,7 +84,10 @@ public class FetchSerializers
{
if (!reply.isOk())
{
out.writeByte(1 + ((CommitOrReadNack) reply).ordinal());
CommitOrReadNack nack = (CommitOrReadNack) reply;
out.writeByte(1 + nack.kind.ordinal());
if (nack.kind == CommitOrReadNack.Kind.InsufficientEpochs)
out.writeUnsignedVInt(nack.minEpoch());
return;
}
@ -100,7 +103,13 @@ public class FetchSerializers
{
int id = in.readByte();
if (id != 0)
return nacks[id - 1];
{
CommitOrReadNack.Kind kind = CommitOrReadNack.Kind.lookupByOrdinal(id - 1);
CommitOrReadNack nack = CommitOrReadNack.lookupByKind(kind);
if (nack != null)
return nack;
return new CommitOrReadNack(kind, in.readUnsignedVInt());
}
return new FetchResponse(deserializeNullable(in, KeySerializers.ranges),
deserializeNullable(in, streamDataSerializer),
@ -111,7 +120,13 @@ public class FetchSerializers
public long serializedSize(ReadReply reply)
{
if (!reply.isOk())
return TypeSizes.BYTE_SIZE;
{
long size = 1;
CommitOrReadNack nack = (CommitOrReadNack) reply;
if (nack.kind == CommitOrReadNack.Kind.InsufficientEpochs)
size += VIntCoding.sizeOfUnsignedVInt(nack.minEpoch());
return size;
}
FetchResponse response = (FetchResponse) reply;
return TypeSizes.BYTE_SIZE

19
src/java/org/apache/cassandra/service/accord/serializers/ReadDataSerializer.java
View File

@ -293,7 +293,6 @@ public class ReadDataSerializer implements IVersionedSerializer<ReadData>
public static final class ReplySerializer<D extends Data> implements IVersionedSerializer<ReadReply>
{
final CommitOrReadNack[] nacks = CommitOrReadNack.values();
private final VersionedSerializer<D, Version> dataSerializer;
public ReplySerializer(VersionedSerializer<D, Version> dataSerializer)
@ -306,7 +305,10 @@ public class ReadDataSerializer implements IVersionedSerializer<ReadData>
{
if (!reply.isOk())
{
out.writeByte(3 + ((CommitOrReadNack) reply).ordinal());
CommitOrReadNack nack = (CommitOrReadNack) reply;
out.writeByte(3 + nack.kind.ordinal());
if (nack.kind == CommitOrReadNack.Kind.InsufficientEpochs)
out.writeUnsignedVInt(nack.minEpoch());
return;
}
@ -327,7 +329,13 @@ public class ReadDataSerializer implements IVersionedSerializer<ReadData>
{
int flags = in.readByte();
if (flags > 2)
return nacks[flags - 3];
{
CommitOrReadNack.Kind kind = CommitOrReadNack.Kind.lookupByOrdinal(flags - 3);
CommitOrReadNack nack = CommitOrReadNack.lookupByKind(kind);
if (nack != null)
return nack;
return new CommitOrReadNack(kind, in.readUnsignedVInt());
}
Ranges unavailable = deserializeNullable(in, KeySerializers.ranges);
D data = dataSerializer.deserialize(in, version);
@ -342,7 +350,12 @@ public class ReadDataSerializer implements IVersionedSerializer<ReadData>
public long serializedSize(ReadReply reply, Version version)
{
if (!reply.isOk())
{
CommitOrReadNack nack = (CommitOrReadNack) reply;
if (nack.kind == CommitOrReadNack.Kind.InsufficientEpochs)
return TypeSizes.BYTE_SIZE + VIntCoding.computeUnsignedVIntSize(nack.minEpoch());
return TypeSizes.BYTE_SIZE;
}
ReadOk readOk = (ReadOk) reply;
long size = TypeSizes.BYTE_SIZE

24
src/java/org/apache/cassandra/service/accord/serializers/TopologySerializers.java
View File

@ -27,7 +27,7 @@ import accord.local.Node;
import accord.primitives.Range;
import accord.topology.Shard;
import accord.topology.Topology;
import accord.utils.SimpleBitSet;
import accord.utils.LargeBitSet;
import accord.utils.SortedArrays;
import accord.utils.SortedArrays.SortedArrayList;
import accord.utils.TinyEnumSet;
@ -41,7 +41,7 @@ import org.apache.cassandra.schema.TableId;
import org.apache.cassandra.service.accord.TokenRange;
import org.apache.cassandra.utils.ArraySerializers;
import org.apache.cassandra.utils.CollectionSerializers;
import org.apache.cassandra.utils.SimpleBitSetSerializer;
import org.apache.cassandra.utils.LargeBitSetSerializer;
import static accord.utils.SortedArrays.fromSimpleBitSet;
@ -220,10 +220,10 @@ public class TopologySerializers
out.writeUnsignedVInt32(rangeIdx);
CollectionSerializers.serializeList(shard.nodes, out, TopologySerializers.nodeId);
SimpleBitSet notInFastPath = SortedArrays.toSimpleBitSet(shard.nodes, shard.notInFastPath);
SimpleBitSetSerializer.instance.serialize(notInFastPath, out);
SimpleBitSet joining = SortedArrays.toSimpleBitSet(shard.nodes, shard.joining);
SimpleBitSetSerializer.instance.serialize(joining, out);
LargeBitSet notInFastPath = SortedArrays.toLargeBitSet(shard.nodes, shard.notInFastPath);
LargeBitSetSerializer.instance.serialize(notInFastPath, out);
LargeBitSet joining = SortedArrays.toLargeBitSet(shard.nodes, shard.joining);
LargeBitSetSerializer.instance.serialize(joining, out);
out.writeUnsignedVInt32(shard.flags().bitset());
}
}
@ -271,10 +271,10 @@ public class TopologySerializers
size += TypeSizes.sizeofUnsignedVInt(rangeIdx);
size += CollectionSerializers.serializedListSize(shard.nodes, TopologySerializers.nodeId);
SimpleBitSet notInFastPath = SortedArrays.toSimpleBitSet(shard.nodes, shard.notInFastPath);
size += SimpleBitSetSerializer.instance.serializedSize(notInFastPath);
SimpleBitSet joining = SortedArrays.toSimpleBitSet(shard.nodes, shard.joining);
size += SimpleBitSetSerializer.instance.serializedSize(joining);
LargeBitSet notInFastPath = SortedArrays.toLargeBitSet(shard.nodes, shard.notInFastPath);
size += LargeBitSetSerializer.instance.serializedSize(notInFastPath);
LargeBitSet joining = SortedArrays.toLargeBitSet(shard.nodes, shard.joining);
size += LargeBitSetSerializer.instance.serializedSize(joining);
size += TypeSizes.sizeofUnsignedVInt(shard.flags().bitset());
}
return size;
@ -301,8 +301,8 @@ public class TopologySerializers
TokenRange range = ranges.get(rangeIndex).withTable(activeTableId);
SortedArrays.SortedArrayList<Node.Id> nodes = CollectionSerializers.deserializeSortedArrayList(in, TopologySerializers.nodeId, Node.Id[]::new);
SimpleBitSet notInFastPath = SimpleBitSetSerializer.instance.deserialize(in);
SimpleBitSet joining = SimpleBitSetSerializer.instance.deserialize(in);
LargeBitSet notInFastPath = LargeBitSetSerializer.instance.deserialize(in);
LargeBitSet joining = LargeBitSetSerializer.instance.deserialize(in);
int flags = in.readUnsignedVInt32();
shards[i] = Shard.SerializerSupport.create(range, nodes, fromSimpleBitSet(nodes, notInFastPath, Node.Id[]::new), fromSimpleBitSet(nodes, joining, Node.Id[]::new), new TinyEnumSet<>(flags));
}

10
src/java/org/apache/cassandra/service/accord/serializers/WaitingOnSerializer.java
View File

@ -30,7 +30,7 @@ import accord.primitives.Timestamp;
import accord.primitives.TxnId;
import accord.utils.ImmutableBitSet;
import accord.utils.Invariants;
import accord.utils.SimpleBitSet;
import accord.utils.LargeBitSet;
import org.apache.cassandra.db.TypeSizes;
import org.apache.cassandra.io.util.DataInputPlus;
import org.apache.cassandra.io.util.DataOutputPlus;
@ -121,9 +121,9 @@ public class WaitingOnSerializer
}
}
private static void serialize(int length, SimpleBitSet write, DataOutputPlus out) throws IOException
private static void serialize(int length, LargeBitSet write, DataOutputPlus out) throws IOException
{
long[] bits = SimpleBitSet.SerializationSupport.getArray(write);
long[] bits = LargeBitSet.SerializationSupport.getArray(write);
Invariants.require(length == bits.length);
for (int i = 0; i < length; i++)
out.writeLong(bits[i]);
@ -137,9 +137,9 @@ public class WaitingOnSerializer
return ImmutableBitSet.SerializationSupport.construct(bits);
}
public static long serializedSize(int length, SimpleBitSet write)
public static long serializedSize(int length, LargeBitSet write)
{
long[] bits = SimpleBitSet.SerializationSupport.getArray(write);
long[] bits = LargeBitSet.SerializationSupport.getArray(write);
Invariants.require(length == bits.length, "Expected length %d != %d", length, bits.length);
return (long) TypeSizes.LONG_SIZE * length;
}

9
src/java/org/apache/cassandra/service/accord/txn/TxnWrite.java
View File

@ -42,7 +42,6 @@ import accord.primitives.Seekable;
import accord.primitives.Seekables;
import accord.primitives.Timestamp;
import accord.primitives.TxnId;
import accord.primitives.Writes;
import accord.utils.async.AsyncChain;
import accord.utils.async.AsyncChains;
import org.apache.cassandra.cql3.UpdateParameters;
@ -456,7 +455,7 @@ public class TxnWrite extends AbstractKeySorted<TxnWrite.Update> implements Writ
// Accord should skip the Update for a read transaction, but handle it here anyways
TxnUpdate txnUpdate = ((TxnUpdate)txn.update());
if (txnUpdate == null)
return Writes.SUCCESS;
return AsyncChains.success(null);
long timestamp = executeAt.uniqueHlc();
@ -477,12 +476,12 @@ public class TxnWrite extends AbstractKeySorted<TxnWrite.Update> implements Writ
}
if (results.isEmpty())
return Writes.SUCCESS;
return AsyncChains.success(null);
if (results.size() == 1)
return results.get(0).flatMap(o -> Writes.SUCCESS);
return results.get(0).map(o -> null);
return AsyncChains.reduce(results, (i1, i2) -> null, (Void)null).flatMap(ignore -> Writes.SUCCESS);
return AsyncChains.reduce(results, (i1, i2) -> null, (Void)null).map(ignore -> null);
}
public long estimatedSizeOnHeap()

5
src/java/org/apache/cassandra/service/paxos/ContentionStrategy.java
View File

@ -333,6 +333,11 @@ public class ContentionStrategy extends RetryStrategy
}
}
public long getMaxMicros(int attempts)
{
return max(min, min(max, delegate.getMaxMicros(attempts)));
}
public String toString()
{
return "Bound{" +

3
src/java/org/apache/cassandra/tcm/RemoteProcessor.java
View File

@ -133,12 +133,13 @@ public final class RemoteProcessor implements Processor
if (waitFor == null)
return fetchLogAndWait(new CandidateIterator(candidates(true), false), log);
Future<ClusterMetadata> cmFuture = null;
try
{
Supplier<Future<ClusterMetadata>> fetchFunction = () -> fetchLogAndWaitInternal(new CandidateIterator(candidates(true), false),
log);
Future<ClusterMetadata> cmFuture = EpochAwareDebounce.instance.getAsync(fetchFunction, waitFor);
cmFuture = EpochAwareDebounce.instance.getAsync(fetchFunction, waitFor);
return cmFuture.get(retryPolicy.remainingNanos(), TimeUnit.NANOSECONDS);
}
catch (InterruptedException e)

26
src/java/org/apache/cassandra/transport/ClientResourceLimits.java
View File

@ -24,11 +24,12 @@ import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.atomic.AtomicInteger;
import com.google.common.annotations.VisibleForTesting;
import org.apache.cassandra.metrics.CassandraReservoir;
import org.apache.cassandra.utils.concurrent.NonBlockingRateLimiter;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import com.codahale.metrics.Reservoir;
import com.codahale.metrics.Snapshot;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.metrics.DecayingEstimatedHistogramReservoir;
@ -117,24 +118,17 @@ public class ClientResourceLimits
* This will recompute the ip usage histo on each query of the snapshot when requested instead of trying to keep
* a histogram up to date with each request
*/
public static Reservoir ipUsageReservoir()
public static CassandraReservoir ipUsageReservoir()
{
return new Reservoir()
return new CassandraReservoir()
{
public int size()
{
return PER_ENDPOINT_ALLOCATORS.size();
}
@Override public Snapshot getPercentileSnapshot() { throw new UnsupportedOperationException(); }
@Override public long[] buckets(int length) { throw new UnsupportedOperationException(); }
@Override public BucketStrategy bucketStrategy() { return BucketStrategy.none; }
@Override public void update(long l) { throw new UnsupportedOperationException(); }
public void update(long l)
{
throw new IllegalStateException();
}
public Snapshot getSnapshot()
{
return getCurrentIpUsage();
}
@Override public int size() { return PER_ENDPOINT_ALLOCATORS.size(); }
@Override public Snapshot getSnapshot() { return getCurrentIpUsage(); }
};
}

15
src/java/org/apache/cassandra/utils/EstimatedHistogram.java
View File

@ -88,6 +88,19 @@ public class EstimatedHistogram implements DoubleToLongFunction
buckets = new AtomicLongArray(bucketData);
}
public static long[] newOffsetsWithScale(long maxScale, boolean considerZeroes)
{
int i = 2;
long next = 1;
while (true)
{
next = Math.max(next + 1, Math.round(next * 1.2));
if (next >= maxScale)
return newOffsets(i, considerZeroes);
++i;
}
}
public static long[] newOffsets(int size, boolean considerZeroes)
{
long[] result = new long[size + (considerZeroes ? 1 : 0)];
@ -256,6 +269,8 @@ public class EstimatedHistogram implements DoubleToLongFunction
sum += bCount * bucketOffsets[i];
}
if (elements == 0)
return 0.0D;
return (double) sum / elements;
}

18
src/java/org/apache/cassandra/utils/SimpleBitSetSerializer.java → src/java/org/apache/cassandra/utils/LargeBitSetSerializer.java
View File

@ -20,20 +20,20 @@ package org.apache.cassandra.utils;
import java.io.IOException;
import accord.utils.SimpleBitSet;
import accord.utils.LargeBitSet;
import org.apache.cassandra.db.TypeSizes;
import org.apache.cassandra.io.UnversionedSerializer;
import org.apache.cassandra.io.util.DataInputPlus;
import org.apache.cassandra.io.util.DataOutputPlus;
public class SimpleBitSetSerializer implements UnversionedSerializer<SimpleBitSet>
public class LargeBitSetSerializer implements UnversionedSerializer<LargeBitSet>
{
public static final SimpleBitSetSerializer instance = new SimpleBitSetSerializer();
public static final LargeBitSetSerializer instance = new LargeBitSetSerializer();
@Override
public void serialize(SimpleBitSet t, DataOutputPlus out) throws IOException
public void serialize(LargeBitSet t, DataOutputPlus out) throws IOException
{
long[] raw = SimpleBitSet.SerializationSupport.getArray(t);
long[] raw = LargeBitSet.SerializationSupport.getArray(t);
// find the first word written
int wordsInUse = wordsInUse(raw);
out.writeUnsignedVInt32(raw.length);
@ -43,20 +43,20 @@ public class SimpleBitSetSerializer implements UnversionedSerializer<SimpleBitSe
}
@Override
public SimpleBitSet deserialize(DataInputPlus in) throws IOException
public LargeBitSet deserialize(DataInputPlus in) throws IOException
{
int size = in.readUnsignedVInt32();
long[] raw = new long[size];
int wordsInUse = in.readUnsignedVInt32();
for (int i = 0; i < wordsInUse; i++)
raw[i] = in.readUnsignedVInt();
return SimpleBitSet.SerializationSupport.construct(raw);
return LargeBitSet.SerializationSupport.construct(raw);
}
@Override
public long serializedSize(SimpleBitSet t)
public long serializedSize(LargeBitSet t)
{
long[] raw = SimpleBitSet.SerializationSupport.getArray(t);
long[] raw = LargeBitSet.SerializationSupport.getArray(t);
// find the last word written
int wordsInUse = wordsInUse(raw);
long size = TypeSizes.sizeofUnsignedVInt(raw.length);

8
test/distributed/org/apache/cassandra/distributed/test/accord/AccordBootstrapTest.java
View File

@ -124,8 +124,8 @@ public class AccordBootstrapTest extends TestBaseImpl
{
AccordConfigurationService configService = service().configService();
boolean completed = configService.unsafeLocalSyncNotified(epoch).await(30, TimeUnit.SECONDS);
Assert.assertTrue(String.format("Local sync notification for epoch %s did not become ready within timeout on %s",
epoch, FBUtilities.getBroadcastAddressAndPort()), completed);
Assert.assertTrue(String.format("Local sync notification for epoch %s did not become ready within timeout on %s\n%s",
epoch, FBUtilities.getBroadcastAddressAndPort(), service().configService().getDebugStr()), completed);
}
catch (InterruptedException e)
{
@ -180,6 +180,8 @@ public class AccordBootstrapTest extends TestBaseImpl
.set("accord.queue_shard_count", 2)
.set("accord.shard_durability_cycle", "20s")
.set("accord.shard_durability_target_splits", "1")
.set("accord.retry_syncpoint", "1s*attempts")
.set("accord.retry_durability", "1s*attempts")
.with(NETWORK, GOSSIP))
.start())
{
@ -190,10 +192,8 @@ public class AccordBootstrapTest extends TestBaseImpl
for (IInvokableInstance node : cluster)
{
node.runOnInstance(() -> {
Assert.assertEquals(initialMax, ClusterMetadata.current().epoch.getEpoch());
System.out.println("Awaiting " + initialMax);
awaitEpoch(initialMax);
AccordConfigurationService configService = service().configService();
long minEpoch = configService.minEpoch();

3
test/distributed/org/apache/cassandra/distributed/test/accord/AccordLoadTest.java
View File

@ -117,7 +117,8 @@ public class AccordLoadTest extends AccordTestBase
final long batchTime = TimeUnit.SECONDS.toNanos(10);
final int concurrency = 100;
final int ratePerSecond = 1000;
final int keyCount = 10_000;
// final int keyCount = 10_000;
final int keyCount = 10;
final float readChance = 0.33f;
long nextRepairAt = repairInterval;
long nextCompactionAt = compactionInterval;

58
test/distributed/org/apache/cassandra/distributed/test/accord/AccordMetricsTest.java
View File

@ -43,7 +43,8 @@ import org.apache.cassandra.distributed.api.Row;
import org.apache.cassandra.distributed.api.SimpleQueryResult;
import org.apache.cassandra.exceptions.ReadTimeoutException;
import org.apache.cassandra.exceptions.WriteTimeoutException;
import org.apache.cassandra.metrics.AccordMetrics;
import org.apache.cassandra.metrics.AccordCoordinatorMetrics;
import org.apache.cassandra.metrics.AccordReplicaMetrics;
import org.apache.cassandra.metrics.DefaultNameFactory;
import org.apache.cassandra.metrics.RatioGaugeSet;
import org.apache.cassandra.net.Verb;
@ -80,7 +81,7 @@ public class AccordMetricsTest extends AccordTestBase
ProtocolModifiers.Toggles.setPermittedFastPaths(new FastPaths(FastPath.Unoptimised));
}));
for (int i = 0; i < SHARED_CLUSTER.size(); i++) // initialize metrics
logger.trace(SHARED_CLUSTER.get(i + 1).callOnInstance(() -> AccordMetrics.readMetrics.toString() + AccordMetrics.writeMetrics.toString()));
logger.trace(SHARED_CLUSTER.get(i + 1).callOnInstance(() -> AccordCoordinatorMetrics.readMetrics.toString() + AccordCoordinatorMetrics.writeMetrics.toString()));
}
String writeCql()
@ -173,9 +174,8 @@ public class AccordMetricsTest extends AccordTestBase
assertClientMetrics(0, "AccordWrite", 1, 0);
assertClientMetrics(1, "AccordWrite", 0, 0);
// TODO (required): reenable or remove metric
// assertCoordinatorMetrics(0, "rw", 0, 0, 1, 0, 0);
assertCoordinatorMetrics(1, "rw", 0, 0, 0, 0, 0);
assertCoordinatorMetrics(0, "rw", 0, 0, 1, 0, 0);
assertCoordinatorMetrics(1, "rw", 0, 0, 0, 0, 1);
assertReplicaMetrics(0, "rw", 0, 0, 0);
assertReplicaMetrics(1, "rw", 0, 0, 0);
@ -195,8 +195,8 @@ public class AccordMetricsTest extends AccordTestBase
assertClientMetrics(0, "AccordRead", 1, 0);
assertClientMetrics(1, "AccordRead", 0, 0);
// TODO (required): reenable or remove metric
// assertCoordinatorMetrics(0, "ro", 0, 0, 1, 0, 0);
assertCoordinatorMetrics(1, "ro", 0, 0, 0, 0, 0);
assertCoordinatorMetrics(0, "ro", 0, 0, 1, 0, 0);
assertCoordinatorMetrics(1, "ro" , 0, 0, 0, 0, 1);
assertReplicaMetrics(0, "ro", 0, 0, 0);
assertReplicaMetrics(1, "ro", 0, 0, 0);
@ -231,7 +231,7 @@ public class AccordMetricsTest extends AccordTestBase
assertClientMetrics(0, "AccordRead", 0, 1);
assertClientMetrics(1, "AccordRead", 0, 0);
// TODO (required): reenable or remove metric
// TODO (required): rework tests: internal accord timeout triggers later than external C* client one
// assertCoordinatorMetrics(0, "ro", 0, 0, 0, 1, 0);
assertCoordinatorMetrics(1, "ro", 0, 0, 0, 0, 0);
assertReplicaMetrics(0, "ro", 0, 0, 0);
@ -252,13 +252,15 @@ public class AccordMetricsTest extends AccordTestBase
assertClientMetrics(0, "AccordWrite", 0, 1);
assertClientMetrics(1, "AccordWrite", 0, 0);
// TODO (required): reenable or remove metric
// assertCoordinatorMetrics(0, "rw", 0, 0, 0, 1, 0);
assertCoordinatorMetrics(0, "rw", 0, 0, 0, 1, 0);
assertCoordinatorMetrics(1, "rw", 0, 0, 0, 0, 0);
assertReplicaMetrics(0, "rw", 0, 0, 0);
assertReplicaMetrics(1, "rw", 0, 0, 0);
assertZeroMetrics("ro","AccordRead");
assertCoordinatorMetrics(0, "ro", 0, 0, 0, 0, 1);
assertCoordinatorMetrics(1, "ro", 0, 0, 0, 0, 0);
assertReplicaMetrics(0, "ro", 0, 0, 0);
assertReplicaMetrics(1, "ro", 0, 0, 0);
}
private void assertZeroMetrics(String scope, String clientScope)
@ -293,17 +295,17 @@ public class AccordMetricsTest extends AccordTestBase
private void assertCoordinatorMetrics(int node, String scope, long fastPaths, long slowPaths, long preempts, long timeouts, long recoveries)
{
DefaultNameFactory nameFactory = new DefaultNameFactory(AccordMetrics.ACCORD_COORDINATOR, scope);
DefaultNameFactory nameFactory = new DefaultNameFactory(AccordCoordinatorMetrics.ACCORD_COORDINATOR, scope);
Map<String, Long> metrics = diff(countingMetrics0).get(node);
logger.info("Metrics for node {} / {}: {}", node, scope, metrics);
Function<String, Long> metric = n -> metrics.get(nameFactory.createMetricName(n).getMetricName());
assertThat(metric.apply(AccordMetrics.FAST_PATHS)).isEqualTo(fastPaths);
assertThat(metric.apply(AccordMetrics.SLOW_PATHS)).isEqualTo(slowPaths);
assertThat(metric.apply(AccordMetrics.PREEMPTED)).isEqualTo(preempts);
assertThat(metric.apply(AccordMetrics.TIMEOUTS)).isEqualTo(timeouts);
assertThat(metric.apply(AccordMetrics.RECOVERY_DELAY)).isEqualTo(recoveries);
assertThat(metric.apply(AccordMetrics.RECOVERY_TIME)).isEqualTo(recoveries);
assertThat(metric.apply(AccordMetrics.COORDINATOR_DEPENDENCIES)).isEqualTo(fastPaths + slowPaths);
assertThat(metric.apply(AccordCoordinatorMetrics.FAST_PATHS)).isEqualTo(fastPaths);
assertThat(metric.apply(AccordCoordinatorMetrics.SLOW_PATHS)).isEqualTo(slowPaths);
assertThat(metric.apply(AccordCoordinatorMetrics.PREEMPTED)).isEqualTo(preempts);
assertThat(metric.apply(AccordCoordinatorMetrics.TIMEOUTS)).isEqualTo(timeouts);
assertThat(metric.apply(AccordCoordinatorMetrics.RECOVERY_DELAY)).isEqualTo(recoveries);
assertThat(metric.apply(AccordCoordinatorMetrics.RECOVERY_TIME)).isEqualTo(recoveries);
assertThat(metric.apply(AccordCoordinatorMetrics.COORDINATOR_DEPENDENCIES)).isEqualTo(fastPaths + slowPaths);
// Verify that coordinator metrics are published to the appropriate virtual table:
SimpleQueryResult res = SHARED_CLUSTER.get(node + 1)
@ -317,21 +319,21 @@ public class AccordMetricsTest extends AccordTestBase
if ((fastPaths + slowPaths) > 0)
{
String fastPathToTotalName = nameFactory.createMetricName(AccordMetrics.FAST_PATH_TO_TOTAL + "." + RatioGaugeSet.MEAN_RATIO).getMetricName();
String fastPathToTotalName = nameFactory.createMetricName(AccordCoordinatorMetrics.FAST_PATH_TO_TOTAL + "." + RatioGaugeSet.MEAN_RATIO).getMetricName();
assertThat((double) SHARED_CLUSTER.get(1).metrics().getGauge(fastPathToTotalName)).isEqualTo((double) fastPaths / (double) (fastPaths + slowPaths), Offset.offset(0.01d));
}
}
private void assertReplicaMetrics(int node, String scope, long stable, long executions, long applications)
{
DefaultNameFactory nameFactory = new DefaultNameFactory(AccordMetrics.ACCORD_REPLICA, scope);
DefaultNameFactory nameFactory = new DefaultNameFactory(AccordReplicaMetrics.ACCORD_REPLICA, scope);
Map<String, Long> metrics = diff(countingMetrics0).get(node);
Function<String, Long> metric = n -> metrics.get(nameFactory.createMetricName(n).getMetricName());
assertThat(metric.apply(AccordMetrics.REPLICA_STABLE_LATENCY)).isLessThanOrEqualTo(stable);
assertThat(metric.apply(AccordMetrics.REPLICA_PREAPPLY_LATENCY)).isEqualTo(executions);
assertThat(metric.apply(AccordMetrics.REPLICA_APPLY_LATENCY)).isEqualTo(applications);
assertThat(metric.apply(AccordMetrics.REPLICA_APPLY_DURATION)).isEqualTo(scope.equals("rw") ? applications : 0);
assertThat(metric.apply(AccordMetrics.REPLICA_DEPENDENCIES)).isEqualTo(executions);
assertThat(metric.apply(AccordReplicaMetrics.REPLICA_STABLE_LATENCY)).isLessThanOrEqualTo(stable);
assertThat(metric.apply(AccordReplicaMetrics.REPLICA_PREAPPLY_LATENCY)).isEqualTo(executions);
assertThat(metric.apply(AccordReplicaMetrics.REPLICA_APPLY_LATENCY)).isEqualTo(applications);
assertThat(metric.apply(AccordReplicaMetrics.REPLICA_APPLY_DURATION)).isEqualTo(scope.equals("rw") ? applications : 0);
assertThat(metric.apply(AccordReplicaMetrics.REPLICA_DEPENDENCIES)).isEqualTo(executions);
// Verify that replica metrics are published to the appropriate virtual table:
SimpleQueryResult vtableResults = SHARED_CLUSTER.get(node + 1)
@ -356,8 +358,8 @@ public class AccordMetricsTest extends AccordTestBase
for (int i = 0; i < SHARED_CLUSTER.size(); i++)
{
Map<String, Long> map = SHARED_CLUSTER.get(i + 1).metrics().getCounters(name -> name.startsWith("org.apache.cassandra.metrics.Accord") || (name.startsWith("org.apache.cassandra.metrics.ClientRequest") && (name.endsWith("AccordRead") || name.endsWith("AccordWrite"))));
SHARED_CLUSTER.get(i + 1).metrics().getGauges(name -> name.startsWith("org.apache.cassandra.metrics.AccordReplica"))
.forEach((key, value) -> map.put(key, (Long)value));
SHARED_CLUSTER.get(i + 1).metrics().getGauges(name -> name.startsWith("org.apache.cassandra.metrics.Accord"))
.forEach((key, value) -> map.put(key, ((Number)value).longValue()));
metrics.put(i, map);
}
return metrics;

11
test/distributed/org/apache/cassandra/distributed/test/accord/AccordMigrationReadRaceTestBase.java
View File

@ -36,6 +36,7 @@ import java.util.concurrent.ExecutionException;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.function.Consumer;
import java.util.stream.Collectors;
import com.google.common.base.Stopwatch;
import com.google.common.collect.ImmutableList;
@ -47,7 +48,10 @@ import org.junit.Test;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import accord.coordinate.Coordination;
import accord.coordinate.Coordination.CoordinationKind;
import accord.primitives.Ranges;
import accord.utils.TinyEnumSet;
import org.apache.cassandra.ServerTestUtils;
import org.apache.cassandra.Util;
import org.apache.cassandra.config.CassandraRelevantProperties;
@ -493,8 +497,11 @@ public abstract class AccordMigrationReadRaceTestBase extends AccordTestBase
// Accord will block until we unpause enactment so to test the routing we wait until the transaction
// has started so the epoch it is created in is the old one
Util.spinUntilTrue(() -> outOfSyncInstance.callOnInstance(() -> {
logger.info("Coordinating {}", AccordService.instance().node().coordinating());
return AccordService.instance().node().coordinating().size() == expectedTransactions;
logger.info("Fetching coordinations...");
TinyEnumSet<CoordinationKind> txnKinds = TinyEnumSet.of(CoordinationKind.PreAccept, CoordinationKind.Propose, CoordinationKind.Stabilise, CoordinationKind.Execute, CoordinationKind.Persist, CoordinationKind.BeginRecovery);
List<Coordination> coordinations = AccordService.instance().node().coordinations().stream().filter(c -> txnKinds.test(c.kind())).collect(Collectors.toList());
logger.info("Coordinating {}", coordinations);
return coordinations.size() == expectedTransactions;
}));
logger.info("Accord node is now coordinating something, unpausing so it can continue to execute");

30
test/distributed/org/apache/cassandra/distributed/test/accord/AccordMigrationWriteRaceTestBase.java
View File

@ -21,7 +21,6 @@ package org.apache.cassandra.distributed.test.accord;
import java.io.IOException;
import java.net.InetSocketAddress;
import java.util.List;
import java.util.Map;
import java.util.Queue;
import java.util.Set;
import java.util.concurrent.Callable;
@ -30,6 +29,7 @@ import java.util.concurrent.ExecutionException;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.function.Consumer;
import java.util.stream.Collectors;
import com.google.common.base.Stopwatch;
import com.google.common.collect.ImmutableList;
@ -43,14 +43,14 @@ import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import accord.api.RoutingKey;
import accord.coordinate.Outcome;
import accord.coordinate.Coordination;
import accord.coordinate.Coordination.CoordinationKind;
import accord.messages.PreAccept;
import accord.primitives.KeyRoute;
import accord.primitives.Ranges;
import accord.primitives.Routable.Domain;
import accord.primitives.Route;
import accord.primitives.TxnId;
import accord.utils.async.AsyncResult;
import accord.utils.TinyEnumSet;
import org.apache.cassandra.ServerTestUtils;
import org.apache.cassandra.Util;
import org.apache.cassandra.batchlog.BatchlogManager;
@ -104,6 +104,7 @@ import org.apache.cassandra.utils.FBUtilities;
import org.apache.cassandra.utils.Pair;
import org.apache.cassandra.utils.concurrent.AsyncPromise;
import org.apache.cassandra.utils.concurrent.Promise;
import org.assertj.core.api.Assertions;
import org.eclipse.jetty.util.ConcurrentHashSet;
import static java.lang.String.format;
@ -564,10 +565,11 @@ public abstract class AccordMigrationWriteRaceTestBase extends AccordTestBase
// otherwise it will just be routed straight to non-Accord.
logger.info("Spinning waiting on a transaction");
Util.spinUntilTrue(() -> {
Map<TxnId, AsyncResult<? extends Outcome>> txns = AccordService.instance().node().coordinating();
if (!txns.isEmpty())
TinyEnumSet<CoordinationKind> txnKinds = TinyEnumSet.of(CoordinationKind.PreAccept, CoordinationKind.Propose, CoordinationKind.Stabilise, CoordinationKind.Execute, CoordinationKind.BeginRecovery);
List<Coordination> coordinations = AccordService.instance().node().coordinations().stream().filter(c -> txnKinds.test(c.kind())).collect(Collectors.toList());
if (!coordinations.isEmpty())
{
logger.info("Found txns {}", txns);
logger.info("Found txns {}", coordinations);
return true;
}
return false;
@ -605,20 +607,22 @@ public abstract class AccordMigrationWriteRaceTestBase extends AccordTestBase
// Accord will block until we unpause enactment so to test the routing we wait until the transaction
// has started so the epoch it is created in is the old one
Util.spinUntilTrue(() -> outOfSyncInstance.callOnInstance(() -> {
Map<TxnId, AsyncResult<? extends Outcome>> coordinating = AccordService.instance().node().coordinating();
if (!coordinating.isEmpty())
logger.info("Accord coordinating: " + coordinating);
return !coordinating.isEmpty();
TinyEnumSet<CoordinationKind> txnKinds = TinyEnumSet.of(CoordinationKind.PreAccept, CoordinationKind.Propose, CoordinationKind.Stabilise, CoordinationKind.Execute, CoordinationKind.BeginRecovery);
List<Coordination> coordinations = AccordService.instance().node().coordinations().stream().filter(c -> txnKinds.test(c.kind())).collect(Collectors.toList());
if (!coordinations.isEmpty())
logger.info("Accord coordinating: " + coordinations);
return !coordinations.isEmpty();
}), 20);
boolean failed = false;
try
{
validation.accept(cluster);
throw new AssertionError("Expected validation to fail");
}
catch (AssertionError e)
{
//ignored
failed = true;
}
Assertions.assertThat(failed).isTrue();
}
else
{

3
test/distributed/org/apache/cassandra/distributed/test/jmx/JMXTestsUtil.java
View File

@ -66,7 +66,8 @@ public class JMXTestsUtil
"org.apache.cassandra.db:type=StorageService:abortMove", // throws since there is no move in progress
"org.apache.cassandra.db:type=CIDRGroupsMappingManager:loadCidrGroupsCache", // AllowAllCIDRAuthorizer doesn't support this operation, as feature is disabled by default
"org.apache.cassandra.db:type=StorageService:forceRemoveCompletion", // deprecated (TCM)
"org.apache.cassandra.db:type=StorageService:createEpochUnsafe" // for Accord testing, but will likely be removed
"org.apache.cassandra.db:type=StorageService:createEpochUnsafe", // for Accord testing, but will likely be removed
"org.apache.cassandra.metrics:type=Client,name=RequestsSizeByIpDistribution:bucketsId" // for Accord testing, but will likely be removed
);
// This set of mbeans are registered early enough during the startup of a
// Cassandra instance for in-jvm dtests to avoid missing registration of mbeans.

2
test/distributed/org/apache/cassandra/service/accord/AccordJournalBurnTest.java
View File

@ -180,7 +180,7 @@ public class AccordJournalBurnTest extends BurnTestBase
5 + random.nextInt(15),
operations,
10 + random.nextInt(30),
new RandomDelayQueue.Factory(random).get(),
RandomDelayQueue::new,
(nodeId, randomSource) -> {
try
{

60
test/unit/org/apache/cassandra/metrics/HitRateTest.java Normal file
View File

@ -0,0 +1,60 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.metrics;
import org.junit.Assert;
import org.junit.Test;
import org.agrona.concurrent.NoOpLock;
public class HitRateTest
{
static long M = 1024;
@Test
public void testRollingCounters()
{
ShardedHitRate.HitRateShard shard = new ShardedHitRate.HitRateShard(new NoOpLock());
ShardedHitRate.Snapshot snapshot = new ShardedHitRate.Snapshot();
long now = M/2;
shard.markHitExclusive(now);
shard.markHitExclusive(now);
shard.markHitExclusive(now += M);
shard.markHitExclusive(now);
shard.markHitExclusive(now += M);
shard.markHitExclusive(now);
shard.markHitExclusive(now += M);
shard.markHitExclusive(now);
shard.markHitExclusive(now += M);
shard.markHitExclusive(now);
shard.updateSnapshot(snapshot, now);
Assert.assertEquals(10, snapshot.totalHits);
Assert.assertEquals(3, snapshot.requests(1));
Assert.assertEquals(9, snapshot.requests(4));
Assert.assertEquals(10, snapshot.requests(5));
Assert.assertEquals(10, snapshot.requests(6));
snapshot.reset();
shard.updateSnapshot(snapshot, now + M/2);
Assert.assertEquals(10, snapshot.totalHits);
Assert.assertEquals(2, snapshot.requests(1));
Assert.assertEquals(8, snapshot.requests(4));
Assert.assertEquals(10, snapshot.requests(5));
Assert.assertEquals(10, snapshot.requests(6));
}
}

2
test/unit/org/apache/cassandra/metrics/JmxVirtualTableMetricsTest.java
View File

@ -67,7 +67,7 @@ public class JmxVirtualTableMetricsTest extends CQLTester
metricToNameMap.put(MetricType.METER, registry.meter("meter"));
metricToNameMap.put(MetricType.COUNTER, registry.counter("counter"));
metricToNameMap.put(MetricType.HISTOGRAM, registry.histogram("histogram"));
metricToNameMap.put(MetricType.HISTOGRAM, registry.histogram("histogram", () -> new CassandraHistogram(new DecayingEstimatedHistogramReservoir(true))));
metricToNameMap.put(MetricType.TIMER, registry.timer("timer"));
metricToNameMap.put(MetricType.GAUGE, registry.gauge("gauge", () -> gaugeValue::get));

147
test/unit/org/apache/cassandra/service/accord/AccordCacheTest.java
View File

@ -17,18 +17,16 @@
*/
package org.apache.cassandra.service.accord;
import java.util.UUID;
import java.util.function.Function;
import org.junit.Assert;
import org.junit.Before;
import org.junit.Test;
import org.agrona.concurrent.NoOpLock;
import org.apache.cassandra.cache.CacheSize;
import org.apache.cassandra.concurrent.ExecutorPlus;
import org.apache.cassandra.concurrent.ManualExecutor;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import org.apache.cassandra.metrics.CacheAccessMetrics;
import org.apache.cassandra.service.accord.AccordCacheEntry.SaveExecutor;
import org.apache.cassandra.service.accord.AccordCacheEntry.Status;
@ -41,7 +39,6 @@ import static org.mockito.Mockito.when;
public class AccordCacheTest
{
private static final long DEFAULT_NODE_SIZE = nodeSize(0);
private AccordCacheMetrics cacheMetrics;
private static abstract class TestSafeState<T> implements AccordSafeState<T, T>
{
@ -134,6 +131,12 @@ public class AccordCacheTest
return itemSize + emptyNodeSize();
}
private static int nextMetricId;
private static String nextMetricId()
{
return Integer.toString(++nextMetricId);
}
private static void assertCacheState(AccordCache cache, int referenced, int total, long bytes)
{
Assert.assertEquals(referenced, cache.numReferencedEntries());
@ -141,33 +144,25 @@ public class AccordCacheTest
Assert.assertEquals(bytes, cache.weightedSize());
}
private void assertCacheMetrics(CacheAccessMetrics metrics, int hits, int misses, int requests, int sizes)
private void assertCacheMetrics(AccordCacheMetrics metrics, int hits, int misses, int requests, int sizes)
{
Assert.assertEquals(hits, metrics.hits.getCount());
Assert.assertEquals(misses, metrics.misses.getCount());
Assert.assertEquals(requests, metrics.requests.getCount());
if (metrics instanceof AccordCacheMetrics)
{
AccordCacheMetrics ascMetrics = (AccordCacheMetrics) metrics;
Assert.assertEquals(sizes, ascMetrics.objectSize.getCount());
assertThat(ascMetrics.objectSize.getSnapshot().getMax()).isGreaterThanOrEqualTo(DEFAULT_NODE_SIZE);
}
metrics.hitRate.refresh();
Assert.assertEquals(hits, metrics.hits.getValue().intValue());
Assert.assertEquals(misses, metrics.misses.getValue().intValue());
Assert.assertEquals(requests, metrics.requests.getValue().intValue());
Assert.assertEquals(sizes, metrics.objectSize.getCount());
}
@Before
public void before()
{
String type = String.format("%s-%s", AccordCommandStores.ACCORD_STATE_CACHE, UUID.randomUUID());
cacheMetrics = new AccordCacheMetrics(type);
}
@Test
public void testAcquisitionAndRelease()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), 500, cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), 500);
AccordCache.Type<String, String, SafeString> type =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
AccordCache.Type<String, String, SafeString>.Instance instance = type.newInstance(null);
assertCacheState(cache, 0, 0, 0);
@ -190,20 +185,23 @@ public class AccordCacheTest
Assert.assertSame(safeString1.global, cache.head());
Assert.assertSame(safeString2.global, cache.tail());
assertCacheMetrics(cache.metrics, 0, 2, 2, 2);
assertCacheMetrics(type.typeMetrics, 0, 2, 2, 2);
assertCacheMetrics(cacheMetrics, 0, 2, 2, 2);
}
@Test
public void testCachingMetricsWithTwoInstances()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
AccordCacheMetrics.Shard shard2 = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), 500, cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), 500);
AccordCache.Type<String, String, SafeString> stringType =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
AccordCache.Type<String, String, SafeString>.Instance stringInstance = stringType.newInstance(null);
AccordCache.Type<Integer, Integer, SafeInt> intType =
cache.newType(Integer.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, ignore -> Integer.BYTES, SafeInt::new);
cache.newType(Integer.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, ignore -> Integer.BYTES, SafeInt::new, shard2);
assertCacheState(cache, 0, 0, 0);
AccordCache.Type<Integer, Integer, SafeInt>.Instance intInstance = intType.newInstance(null);
@ -239,10 +237,13 @@ public class AccordCacheTest
@Test
public void testRotation()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), DEFAULT_NODE_SIZE * 5, cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), DEFAULT_NODE_SIZE * 5);
AccordCache.Type<String, String, SafeString> type =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
assertCacheState(cache, 0, 0, 0);
AccordCache.Type<String, String, SafeString>.Instance instance = type.newInstance(null);
@ -260,15 +261,13 @@ public class AccordCacheTest
Assert.assertSame(items[0].global, cache.head());
Assert.assertSame(items[2].global, cache.tail());
assertCacheState(cache, 0, 3, nodeSize(1) * 3);
assertCacheMetrics(cache.metrics, 0, 3, 3, 3);
assertCacheMetrics(type.typeMetrics, 0, 3, 3, 3);
assertCacheMetrics(cacheMetrics, 0, 3, 3, 3);
SafeString safeString = instance.acquire("1");
Assert.assertEquals(Status.LOADED, safeString.global.status());
assertCacheState(cache, 1, 3, nodeSize(1) * 3);
assertCacheMetrics(cache.metrics, 1, 3, 4, 3);
assertCacheMetrics(type.typeMetrics, 1, 3, 4, 3);
assertCacheMetrics(cacheMetrics, 1, 3, 4, 3);
// releasing item should return it to the tail
instance.release(safeString, null);
@ -280,10 +279,13 @@ public class AccordCacheTest
@Test
public void testEvictionOnAcquire()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), nodeSize(1) * 5, cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), nodeSize(1) * 5);
AccordCache.Type<String, String, SafeString> type =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
AccordCache.Type<String, String, SafeString>.Instance instance = type.newInstance(null);
assertCacheState(cache, 0, 0, 0);
@ -300,8 +302,7 @@ public class AccordCacheTest
assertCacheState(cache, 0, 5, nodeSize(1) * 5);
Assert.assertSame(items[0].global, cache.head());
Assert.assertSame(items[4].global, cache.tail());
assertCacheMetrics(cache.metrics, 0, 5, 5, 5);
assertCacheMetrics(type.typeMetrics, 0, 5, 5, 5);
assertCacheMetrics(cacheMetrics, 0, 5, 5, 5);
SafeString safeString = instance.acquire("5");
Assert.assertTrue(instance.isReferenced(safeString.key()));
@ -312,25 +313,26 @@ public class AccordCacheTest
Assert.assertSame(items[4].global, cache.tail());
Assert.assertFalse(instance.keyIsCached("0", SafeString.class));
Assert.assertFalse(instance.keyIsReferenced("0", SafeString.class));
assertCacheMetrics(cache.metrics, 0, 6, 6, 5);
assertCacheMetrics(type.typeMetrics, 0, 6, 6, 5);
assertCacheMetrics(cacheMetrics, 0, 6, 6, 5);
testLoad(executor, safeString, "5");
instance.release(safeString, null);
assertCacheState(cache, 0, 5, nodeSize(1) * 5);
Assert.assertSame(items[1].global, cache.head());
Assert.assertSame(safeString.global, cache.tail());
assertCacheMetrics(cache.metrics, 0, 6, 6, 6);
assertCacheMetrics(type.typeMetrics, 0, 6, 6, 6);
assertCacheMetrics(cacheMetrics, 0, 6, 6, 5);
}
@Test
public void testEvictionOnRelease()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), nodeSize(1) * 4, cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), nodeSize(1) * 4);
AccordCache.Type<String, String, SafeString> type =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
AccordCache.Type<String, String, SafeString>.Instance instance = type.newInstance(null);
assertCacheState(cache, 0, 0, 0);
@ -344,22 +346,19 @@ public class AccordCacheTest
}
assertCacheState(cache, 5, 5, nodeSize(1) * 5);
assertCacheMetrics(cache.metrics, 0, 5, 5, 5);
assertCacheMetrics(type.typeMetrics, 0, 5, 5, 5);
assertCacheMetrics(cacheMetrics, 0, 5, 5, 5);
Assert.assertNull(cache.head());
Assert.assertNull(cache.tail());
instance.release(items[2], null);
assertCacheState(cache, 4, 4, nodeSize(1) * 4);
assertCacheMetrics(cache.metrics, 0, 5, 5, 5);
assertCacheMetrics(type.typeMetrics, 0, 5, 5, 5);
assertCacheMetrics(cacheMetrics, 0, 5, 5, 5);
Assert.assertNull(cache.head());
Assert.assertNull(cache.tail());
instance.release(items[4], null);
assertCacheState(cache, 3, 4, nodeSize(1) * 4);
assertCacheMetrics(cache.metrics, 0, 5, 5, 5);
assertCacheMetrics(type.typeMetrics, 0, 5, 5, 5);
assertCacheMetrics(cacheMetrics, 0, 5, 5, 5);
Assert.assertSame(items[4].global, cache.head());
Assert.assertSame(items[4].global, cache.tail());
}
@ -367,18 +366,20 @@ public class AccordCacheTest
@Test
public void testMultiAcquireRelease()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), DEFAULT_NODE_SIZE * 4, cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), DEFAULT_NODE_SIZE * 4);
AccordCache.Type<String, String, SafeString> type =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
AccordCache.Type<String, String, SafeString>.Instance instance = type.newInstance(null);
assertCacheState(cache, 0, 0, 0);
SafeString safeString1 = instance.acquire("0");
testLoad(executor, safeString1, "0");
Assert.assertEquals(Status.LOADED, safeString1.global.status());
assertCacheMetrics(cache.metrics, 0, 1, 1, 1);
assertCacheMetrics(type.typeMetrics, 0, 1, 1, 1);
assertCacheMetrics(cacheMetrics, 0, 1, 1, 1);
Assert.assertEquals(1, instance.references("0", SafeString.class));
assertCacheState(cache, 1, 1, nodeSize(1));
@ -388,8 +389,7 @@ public class AccordCacheTest
Assert.assertEquals(Status.LOADED, safeString1.global.status());
Assert.assertEquals(2, instance.references("0", SafeString.class));
assertCacheState(cache, 1, 1, nodeSize(1));
assertCacheMetrics(cache.metrics, 1, 1, 2, 1);
assertCacheMetrics(type.typeMetrics, 1, 1, 2, 1);
assertCacheMetrics(cacheMetrics, 1, 1, 2, 1);
instance.release(safeString1, null);
assertCacheState(cache, 1, 1, nodeSize(1));
@ -400,10 +400,13 @@ public class AccordCacheTest
@Test
public void evictionBlockedOnSaving()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), nodeSize(1) * 3 + nodeSize(3), cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), nodeSize(1) * 3 + nodeSize(3));
AccordCache.Type<String, String, SafeString> type =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
AccordCache.Type<String, String, SafeString>.Instance instance = type.newInstance(null);
assertCacheState(cache, 0, 0, 0);
@ -422,8 +425,7 @@ public class AccordCacheTest
}
assertCacheState(cache, 0, 4, nodeSize(1) * 3 + nodeSize(2));
assertCacheMetrics(cache.metrics, 0, 4, 4, 5);
assertCacheMetrics(type.typeMetrics, 0, 4, 4, 5);
assertCacheMetrics(cacheMetrics, 0, 4, 4, 4);
// force cache eviction
instance.acquire(Integer.toString(0));
@ -441,10 +443,13 @@ public class AccordCacheTest
@Test
public void testUpdates()
{
AccordCacheMetrics cacheMetrics = new AccordCacheMetrics(nextMetricId());
AccordCacheMetrics.Shard shard = cacheMetrics.newShard(new NoOpLock());
ManualExecutor executor = new ManualExecutor();
AccordCache cache = new AccordCache(saveExecutor(executor), 500, cacheMetrics);
AccordCache cache = new AccordCache(saveExecutor(executor), 500);
AccordCache.Type<String, String, SafeString> type =
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new);
cache.newType(String.class, (s, k) -> k, (s, k, c, o) -> null, Function.identity(), (s, k, v) -> true, String::length, SafeString::new, shard);
AccordCache.Type<String, String, SafeString>.Instance instance = type.newInstance(null);
assertCacheState(cache, 0, 0, 0);
@ -463,8 +468,7 @@ public class AccordCacheTest
Assert.assertSame(safeString.global, cache.head());
Assert.assertSame(safeString.global, cache.tail());
assertCacheMetrics(cache.metrics, 0, 1, 1, 2);
assertCacheMetrics(type.typeMetrics, 0, 1, 1, 2);
assertCacheMetrics(cacheMetrics, 0, 1, 1, 1);
}
private CacheSize mockCacheSize(long capacity, long size, int entries)
@ -476,21 +480,6 @@ public class AccordCacheTest
return cacheSize;
}
@Test
public void testAccorStateCacheMetrics()
{
CacheAccessMetrics stringInstance1 = cacheMetrics.forInstance(String.class);
CacheAccessMetrics stringInstance1Dup = cacheMetrics.forInstance(String.class);
CacheAccessMetrics stringInstance2 = cacheMetrics.forInstance(String.class);
CacheAccessMetrics integerInstance1 = cacheMetrics.forInstance(Integer.class);
CacheAccessMetrics integerInstance2 = cacheMetrics.forInstance(Integer.class);
assertThat(stringInstance1).isSameAs(stringInstance1Dup);
assertThat(stringInstance1).isSameAs(stringInstance2);
assertThat(integerInstance1).isSameAs(integerInstance2);
assertThat(stringInstance1).isNotSameAs(integerInstance1);
}
private static SaveExecutor saveExecutor(ExecutorPlus executor)
{
return (saving, identity, save) -> {

4
test/unit/org/apache/cassandra/service/accord/AccordCommandStoreTest.java
View File

@ -49,7 +49,7 @@ import accord.primitives.Txn;
import accord.primitives.TxnId;
import accord.primitives.Writes;
import accord.utils.ImmutableBitSet;
import accord.utils.SimpleBitSet;
import accord.utils.LargeBitSet;
import accord.utils.async.AsyncChains;
import org.apache.cassandra.SchemaLoader;
import org.apache.cassandra.cql3.QueryProcessor;
@ -130,7 +130,7 @@ public class AccordCommandStoreTest
Ballot promised = ballot(1, clock.incrementAndGet(), 1);
Ballot accepted = ballot(1, clock.incrementAndGet(), 1);
Timestamp executeAt = timestamp(1, clock.incrementAndGet(), 1);
SimpleBitSet waitingOnApply = new SimpleBitSet(3);
LargeBitSet waitingOnApply = new LargeBitSet(3);
waitingOnApply.set(1);
Command.WaitingOn waitingOn = new Command.WaitingOn(dependencies.keyDeps.keys(), dependencies.rangeDeps, new ImmutableBitSet(waitingOnApply), new ImmutableBitSet(2));
Pair<Writes, Result> result = AsyncChains.getBlocking(AccordTestUtils.processTxnResult(commandStore, txnId, txn, executeAt));

3
test/unit/org/apache/cassandra/service/accord/AccordTestUtils.java
View File

@ -94,7 +94,6 @@ import org.apache.cassandra.dht.Murmur3Partitioner;
import org.apache.cassandra.dht.Range;
import org.apache.cassandra.dht.Token;
import org.apache.cassandra.io.util.File;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import org.apache.cassandra.schema.Schema;
import org.apache.cassandra.schema.TableId;
import org.apache.cassandra.schema.TableMetadata;
@ -348,7 +347,7 @@ public class AccordTestUtils
public static AccordCommandStore createAccordCommandStore(
Node.Id node, LongSupplier now, Topology topology)
{
AccordExecutor executor = new AccordExecutorSyncSubmit(0, RUN_WITH_LOCK, CommandStore.class.getSimpleName() + '[' + 0 + ']', new AccordCacheMetrics("test"), new AccordAgent());
AccordExecutor executor = new AccordExecutorSyncSubmit(0, RUN_WITH_LOCK, CommandStore.class.getSimpleName() + '[' + 0 + ']', new AccordAgent());
return createAccordCommandStore(node, now, topology, executor);
}

3
test/unit/org/apache/cassandra/service/accord/SimulatedAccordCommandStore.java
View File

@ -89,7 +89,6 @@ import org.apache.cassandra.db.Keyspace;
import org.apache.cassandra.db.compaction.CompactionManager;
import org.apache.cassandra.db.filter.ColumnFilter;
import org.apache.cassandra.db.memtable.Memtable;
import org.apache.cassandra.metrics.AccordCacheMetrics;
import org.apache.cassandra.schema.TableId;
import org.apache.cassandra.service.accord.api.TokenKey;
import org.apache.cassandra.tcm.ClusterMetadata;
@ -273,7 +272,7 @@ public class SimulatedAccordCommandStore implements AutoCloseable
}),
updateHolder,
journal,
new AccordExecutorSimple(0, CommandStore.class.getSimpleName() + '[' + 0 + ']', new AccordCacheMetrics("test"), agent));
new AccordExecutorSimple(0, CommandStore.class.getSimpleName() + '[' + 0 + ']', agent));
this.commandStore.executor().executeDirectlyWithLock(() -> {
commandStore.executor().setCapacity(8 << 20);
commandStore.executor().setWorkingSetSize(4 << 20);

7
test/unit/org/apache/cassandra/service/accord/serializers/CommandsForKeySerializerTest.java
View File

@ -108,7 +108,6 @@ import org.apache.cassandra.dht.Token;
import org.apache.cassandra.schema.KeyspaceParams;
import org.apache.cassandra.schema.TableId;
import org.apache.cassandra.service.StorageService;
import org.apache.cassandra.service.accord.AccordTestUtils;
import org.apache.cassandra.service.accord.api.TokenKey;
import org.apache.cassandra.service.accord.txn.TxnData;
import org.apache.cassandra.service.accord.txn.TxnWrite;
@ -181,7 +180,8 @@ public class CommandsForKeySerializerTest
if (saveStatus.known.isDefinitionKnown())
builder.partialTxn(txn);
builder.setParticipants(StoreParticipants.all(txn.keys().toRoute(txn.keys().get(0).someIntersectingRoutingKey(null))));
StoreParticipants participants = StoreParticipants.all(txn.keys().toRoute(txn.keys().get(0).someIntersectingRoutingKey(null)));
builder.setParticipants(participants);
builder.durability(isDurable ? AllQuorums : NotDurable);
if (saveStatus.known.deps().hasPreAcceptedOrProposedOrDecidedDeps())
{
@ -189,7 +189,7 @@ public class CommandsForKeySerializerTest
{
for (TxnId id : deps)
keyBuilder.add(((Key)txn.keys().get(0)).toUnseekable(), id);
builder.partialDeps(new PartialDeps(AccordTestUtils.fullRange(txn), keyBuilder.build(), RangeDeps.NONE));
builder.partialDeps(new PartialDeps(participants.touches(), keyBuilder.build(), RangeDeps.NONE));
}
}
@ -669,6 +669,7 @@ public class CommandsForKeySerializerTest
@Override public long maxConflictsPruneInterval() { return 0; }
@Override public Txn emptySystemTxn(Kind kind, Routable.Domain domain) { throw new UnsupportedOperationException(); }
@Override public long slowCoordinatorDelay(Node node, SafeCommandStore safeStore, TxnId txnId, TimeUnit units, int retryCount) { return 0; }
@Override public boolean isSlowCoordinator(long elapsed, TimeUnit units, TxnId txnId, int attempt) { return false; }
@Override public long slowReplicaDelay(Node node, SafeCommandStore safeStore, TxnId txnId, int retryCount, ProgressLog.BlockedUntil blockedUntil, TimeUnit units) { return 0; }
@Override public long slowAwaitDelay(Node node, SafeCommandStore safeStore, TxnId txnId, int retryCount, ProgressLog.BlockedUntil retrying, TimeUnit units) { return 0; }
@Override public long retrySyncPointDelay(Node node, int attempt, TimeUnit units) { return 0; }

6
test/unit/org/apache/cassandra/service/accord/serializers/WaitingOnSerializerTest.java
View File

@ -33,7 +33,7 @@ import accord.primitives.RoutingKeys;
import accord.primitives.TxnId;
import accord.utils.Gen;
import accord.utils.Gens;
import accord.utils.SimpleBitSet;
import accord.utils.LargeBitSet;
import accord.utils.Utils;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.dht.Murmur3Partitioner;
@ -95,8 +95,8 @@ public class WaitingOnSerializerTest
int txnIdCount = deps.rangeDeps.txnIdCount();
int keyCount = deps.keyDeps.keys().size();
int[] selected = Gens.arrays(Gens.ints().between(0, txnIdCount + keyCount - 1)).unique().ofSizeBetween(0, txnIdCount + keyCount).next(rs);
SimpleBitSet waitingOn = new SimpleBitSet(txnIdCount + keyCount, false);
SimpleBitSet appliedOrInvalidated = rs.nextBoolean() ? null : new SimpleBitSet(txnIdCount, false);
LargeBitSet waitingOn = new LargeBitSet(txnIdCount + keyCount, false);
LargeBitSet appliedOrInvalidated = rs.nextBoolean() ? null : new LargeBitSet(txnIdCount, false);
for (int i : selected)
{
WaitingOnSets set = appliedOrInvalidated == null || i >= txnIdCount ? WaitingOnSets.APPLY : sets.next(rs);

10
test/unit/org/apache/cassandra/utils/SimpleBitSetSerializerTest.java → test/unit/org/apache/cassandra/utils/LargeBitSetSerializerTest.java
View File

@ -19,27 +19,27 @@
package org.apache.cassandra.utils;
import accord.utils.Gen;
import accord.utils.SimpleBitSet;
import accord.utils.LargeBitSet;
import org.apache.cassandra.io.Serializers;
import org.apache.cassandra.io.util.DataOutputBuffer;
import org.junit.Test;
import static accord.utils.Property.qt;
public class SimpleBitSetSerializerTest
public class LargeBitSetSerializerTest
{
@Test
public void test()
{
@SuppressWarnings({ "resource", "IOResourceOpenedButNotSafelyClosed" }) DataOutputBuffer output = new DataOutputBuffer();
qt().forAll(simpleBitSetGen()).check(bits -> Serializers.testSerde(output, SimpleBitSetSerializer.instance, bits));
qt().forAll(largeBitSetGen()).check(bits -> Serializers.testSerde(output, LargeBitSetSerializer.instance, bits));
}
private static Gen<SimpleBitSet> simpleBitSetGen()
private static Gen<LargeBitSet> largeBitSetGen()
{
return rs -> {
int size = rs.nextInt(0, 1 << 10);
SimpleBitSet bitSet = new SimpleBitSet(size);
LargeBitSet bitSet = new LargeBitSet(size);
if (size == 0 || rs.decide(0.2))
return bitSet; // empty
if (rs.decide(0.2))